Have Inter-Judge Sentencing Disparities Increased in an Advisory

Have Inter-Judge Sentencing Disparities Increased in an AdvisoryGuidelines Regime? Evidence From Booker

Crystal S. Yang∗

The Federal Sentencing Guidelines were promulgated in response to con-cerns of widespread disparities in sentencing. After almost two decades of de-terminate sentencing, the Guidelines were rendered advisory in United Statesv. Booker. What has been the result of reintroducing greater judicial discre-tion on inter-judge disparities? This Article utilizes new data to undertakethe first national empirical analysis of inter-judge disparities post Booker.Relying on the random assignment of cases to judges, I find that inter-judgedisparities have doubled since the Guidelines became advisory. Some of therecent increase in disparities can be attributed to differential sentencing be-havior associated with judge demographic characteristics. The application ofmandatory minimums by prosecutors, potentially through the use of supersed-ing indictments, is another prominent source of disparities.

JEL Classifications: J15, K14, K40

∗Olin Fellow and Instructor in Law, University of Chicago Law School. Support was providedby Harvard Law School’s John M. Olin Center for Law, Economics, and Business and the Multidis-ciplinary Program on Inequality and Social Policy at Harvard. I also thank Gary Chamberlain, RajChetty, Will Dobbie, Judge Nancy Gertner, Claudia Goldin, Paul Hofer, Louis Kaplow, LawrenceKatz, Ilyana Kuziemko, J.J. Prescott, Shayak Sarkar, Steven Shavell, and Kate Stith for helpful com-ments and suggestions, and participants in the Harvard Labor Economics/Public Finance Seminars,Harvard Law School Law and Economics Seminar, and Chicago Junior Faculty Workshop. TheTransactional Records Access Clearinghouse (TRAC) at Syracuse University, in particular SusanLong, generously provided sentencing data for use in this project in my role as a TRAC Fellow ofthe Center. All errors are my own. E-mail: [email protected].

TABLE OF CONTENTS

INTRODUCTION ................................................................................................... 1

I BRIEF LEGAL BACKGROUND OF FEDERAL SENTENCING ........................ 8A. Adoption of the Federal Sentencing Guidelines ................................... 8B. Challenges to the Mandatory Guidelines Regime................................. 12

II FRAMEWORK, DATA, AND METHODS ........................................................ 15A. Judicial Behavior in Criminal Sentencing ............................................ 15B. Sentencing Data .................................................................................... 17C. Matching ............................................................................................... 20D. Testing for Random Assignment........................................................... 20E. Trends in Sentencing............................................................................. 23F. Measuring Inter-Judge Disparity: Analysis of Variance....................... 25

III RESULTS ON INTER-JUDGE AND REGIONAL DISPARITIES ....................... 27A. Sentence Length.................................................................................... 27B. Below Range Departures....................................................................... 30C. Above Range Departures ...................................................................... 33D. Sentencing Practices by Judge Demographics...................................... 35E. Regional Disparity: Inter-District Variation.......................................... 39F. Prosecutorial Contributions to Disparities ............................................ 43

IV POLICY RECOMMENDATIONS ...................................................................... 45A. “Topless” Guidelines............................................................................. 46B. “Blakely-ized” Guidelines..................................................................... 47C. Judge Sessions Proposal........................................................................ 48

V CONCLUSION ................................................................................................ 50

APPENDIX............................................................................................................ 52A. Testing for Random Assignment........................................................... 52B. Analysis of Variance ............................................................................. 56C. Judge Demographic Regression............................................................ 57

Inter-Judge Disparities After Booker 1

INTRODUCTION

The Federal Sentencing Guidelines were adopted to counter widespread disparitiesin federal sentencing. By the 1970s, the federal system exhibited “an unjustifiablywide range of sentences to offenders convicted of similar crimes” because eachjudge was “left to apply his own notions of the purposes of sentencing.”1 Disparitieswere so pronounced that a defendant sentenced to three years by one judge wouldhave been sentenced to twenty years had he been assigned to another judge.2 De-crying these disparities and championing sentencing reform, federal district judgeMarvin Frankel claimed that “the almost wholly unchecked and sweeping powerswe give to judges in the fashioning of sentences are terrifying and intolerable for asociety that professes devotion to the rule of law.”3

In response, policymakers sought to limit the “unfettered discretion the law con-fers on those judges and parole authorities [that implement] the sentence.”4 Underthe Sentencing Reform Act (SRA) of 1984, Congress created the United StatesSentencing Commission to promulgate the Guidelines,5 a new regime intended toreduce disparities stemming from judicial preferences and biases rather than offenseand offender characteristics.6 Congress directed the Commission, an independentagency within the judicial branch,7 to fashion sentencing guidelines aimed at the

1See S. REP. NO. 97-307, at 5 (1981).2Anthony Partridge & William B. Eldridge, Federal Judicial Center, The Second Circuit Study:

A Report to the Judges of the Second Circuit 36 (1974).3Marvin E. Frankel, CRIMINAL SENTENCES: LAW WITHOUT ORDER 5 (1973).4See S. Rep. No. 98-225, at 38 (1983) (Senate Report on precursor to federal Sentencing Reform

Act of 1984) (“[E]very day Federal judges mete out an unjustifiably wide range of sentences tooffenders with similar histories, convicted of similar crimes, committed under similar circumstances.. . . These disparities, whether they occur at the time of the initial sentencing or at the parolestage, can be traced directly to the unfettered discretion the law confers on those judges and paroleauthorities responsible for imposing and implementing the sentence”); Id. at 49 (“[T]he presentpractices of the federal courts and of the parole commission clearly indicate that sentencing in thefederal courts is characterized by unwarranted disparity and by uncertainty about the length of timeoffenders will service in prison.”).

5Pub. L. No. 98-473, 98 Stat 1837, 1987 (1984) (codified as amended at 18 USC §3551 et seq.,28 USC §991 et seq.). For a discussion of the efforts leading up the the promulgation of the SRA,see Kate Stith & Steve Y. Koh, The Politics of Sentencing Reform: The Legislative History of theFederal Sentencing Guidelines, 28 WAKE FOREST L. REV. 223, 230 (1993).

6U.S.S.G. §1A.1, intro to comment., pt. A, ¶2 (Congress “sought uniformity in sentencingby narrowing the wide disparity in sentences imposed by different federal courts for similar crim-inal conduct”); S. REP. NO. 225, at 45 (1983) (“Sentencing disparities that are not justified bydifferences among offenses or offenders are unfair both to offenders and to the public.”).

7See 28 U.S.C. §991(a). The Commission was placed in the Judicial Branch because Congressconcluded that “sentencing should remain primarily a judicial function,” and because sitting judgeswould serve on the Commission. The Commission is comprised of seven voting members. TheSRA provides that “[a]t least three of the [Commission] members shall be Federal judges selectedafter considering a list of six judges recommended to the President by the Judicial Conference ofthe United States” and no more than four members of the Commission could be members of thesame political party. See id. The Commission later withstood separation-of-powers challenges, asthe Court rejected several constitutional challenges to the Commission and its delegated authority.See Mistretta v. United States, 488 U.S. 361 (1989).


primary goal of avoiding unwarranted sentencing disparity.8

After the implementation of the Guidelines, researchers began to investigate theextent to which the Guidelines reduced disparities.9 Initial work by Anderson, Stith,and Kling revealed that the Guidelines were somewhat successful in reducing inter-judge sentencing disparity.10 The authors estimate that the expected difference insentence length between judges fell from 17% (4.9 months) in 1986-1987 to 11%(3.9 months) in 1988-93 in 25 cities where case assignment was found sufficientlyrandom.11 Another study by Hofer, Blackwell, and Ruback also found evidenceof reduced inter-judge sentencing disparities after the promulgation of the Guide-lines.12 The study concludes that the Guidelines achieved “modest success” in re-ducing inter-judge disparities, documenting that the sentencing judge accountedfor 2.32% of the variation in sentences in 1984-1985, but only 1.24% under theGuidelines in 1994-1995.13 Convergence in findings by outside researchers and theCommission led the Commission to conclude that “the federal sentencing guide-lines have made significant progress toward reducing disparity caused by judicialdiscretion.”14

After almost two decades of mandatory Guidelines sentencing, the Guidelineswere struck down in United States v. Booker,15 dramatically altering the sentencinglandscape. In Booker, the Supreme Court found that the Guidelines violated theSixth Amendment right to a jury trial and rendered the Guidelines advisory. Subse-quent Supreme Court decisions furthered weakened the effect of the Guidelines oncriminal sentencing by reducing the degree of appellate scrutiny for both within andoutside Guidelines sentences. In Rita v. United States, the Court directed courts ofappeals to apply a presumption of reasonableness to within Guidelines sentences.16

8See, e.g., Stephen Breyer, The Federal Sentencing Guidelines and the Key Compromises UponWhich They Rest, 17 HOFSTRA L. REV. 1, 3 (1988) (Congress sought to reduce "unjustifiablywide" sentencing disparity.); Kenneth R. Feinberg, Federal Criminal Sentencing Reform: Congressand the United States Sentencing Commission, 28 WAKE FOREST L. REV. 291, 295 (1993) (“Thefirst and foremost goal of the sentencing reform effort was to alleviate the perceived problem offederal criminal sentencing disparity.”).

9While the Guidelines were effectively mandatory, the Guidelines did provide a permissiblerange of sentence lengths. Thus, one should expect that the Guidelines would ameliorate, but notcompletely eliminate, inter-judge disparities.

10James M. Anderson et al., Measuring Interjudge Sentencing Disparity: Before and After theFederal Sentencing Guidelines, 42 J.L. & ECON. 271, 303 (1999) (“The Guidelines have re-duced the net variation in sentence attributable to the happenstance of the identity of the sentencingjudge.”).

11Id. at 294.12Paul J. Hofer, Kevin R. Blackwell, & R. Barry Ruback, The Effect of the Federal Sentencing

Guidelines on Inter-Judge Sentencing Disparity, 90 J. CRIM. L. & CRIMINOLOGY 239 (1999). Seeid. at 284-86 for a discussion of the statistical techniques employed in the study.

13Id. at 241, 287 (“Together with the other research reviewed below, [our] findings suggest thatthe sentencing guidelines have had modest but meaningful success at reducing unwarranted disparityamong judges in the sentences imposed on similar crimes and offenders.”)

14United States Sentencing Commission, FIFTEEN YEARS OF GUIDELINES SENTENCING: ANASSESSMENT OF HOW WELL THE FEDERAL CRIMINAL JUSTICE SYSTEM IS ACHIEVING THEGOALS OF SENTENCING REFORM 99 (Nov. 2004).

15543 U.S. 220 (2005).16551 U.S. 338, 347-50 (2007).


Later in Gall v. United States, the Court held that appellate courts could not presumethat a sentence outside the Guidelines range was unreasonable, reducing the degreeof appellate review to a more deferential abuse of discretion standard.17 Concurrentwith the Gall decision, the Court in Kimbrough held that federal district court judgeshave the discretion to impose sentences outside the recommended Guidelines rangedue to policy disagreements with the Sentencing Commission.18

Following Booker and its progeny, Rita, Gall, Kimbrough, the legal commu-nity expressed concerns on the impact of such decisions on inter-judge sentencingdisparities. Congressman Tom Feeney wrote that “the Supreme’s Court decision[in Booker] to place this extraordinary power to sentence a person solely in thehands of a single federal judge - who is accountable to no one - flies in the face ofthe clear will of Congress.”19 U.S. Attorney for the Northern District of Illinois,Patrick Fitzgerald, stated that Booker has “re-introduced into federal sentencingboth substantial district-to-district variations and substantial judge-to-judge vari-ations.”20 Similarly, scholars commented on the huge increase in the degree ofjudicial discretion afforded to judges,21 and predicted an increase in unwarrantedsentencing disparities.22

Due to suggestive evidence of increasing disparities post Booker, the UnitedStates Sentencing Commission and policymakers have commented on possible waysto constrain judicial discretion. Then Attorney General Alberto Gonzales claimedthat in light of “increasing disparity in sentences” since Booker, the Guidelinesneeded to be fixed.23 As a potential “fix,” former Chair of the Sentencing Commis-sion, Judge William K. Sessions III, has suggested “resurrecting” the mandatoryGuidelines in order to give them greater weight during sentencing.24

17552 U.S. 38, 52-53 (2007).18Kimbrough v. United States, 552 U.S. 85, 101 (2007).19Carl Hulse & Adam Liptak, New Fight Over Controlling Punishments Is Widely Seen, N.Y.

TIMES, Jan. 13, 2005, at A29.20Patrick J. Fitzgerald, Testimony before the U.S. Sentencing Commission Regional Hearing in

Chicago, Illinois, at 3 (Sept. 10, 2009).21See Nancy Gertner, A Short History of American Sentencing: Too Little Law, Too Much Law,

or Just Right, 100 J. CRIM. L. & CRIMINOLOGY 691, 706 (2010) (“It is clear that Booker hasenhanced the position of the judge, whose sentencing expertise has been formally acknowledgedagain, at the cost of diminishing the position of the Sentencing Commission.”); Douglas A. Berman,Foreword: Beyond Blakely and Booker: Pondering Modern Sentencing Process, 95 J. CRIM. L.& CRIMINOLOGY 653, 676 (2005) (“Booker devised a new system of federal sentencing whichgranted judges more sentencing power than they had ever previously wielded”); Luiza Ch. Savage,Chaos Ahead After Sentencing Guidelines Decision, N.Y. SUN, Jan. 13, 2005, at 1 (quoting FrankBowman) (arguing that Booker resulted in “the most amount of judicial discretion ever afforded tosentencing judges”).

22See Frank O. Bowman, III, Debacle: How the Supreme Court Has Mangled American Sen-tencing Law and How It Might Yet Be Mended, 77 U. CHI. L. REV. 367, 470 (2010) (“Kimbrough,Gall ... have so thoroughly denatured appellate review that the federal system’s ability to controlregional and judge-to-judge sentencing disparity has been effectively eliminated”).

23Federal Sentencing Guidelines Speech (June 21, 2005), 17 FED. SENTENCING REP. 324,325-26 (2005).

24William K. Sessions III, At the Crossroads of the Three Branches: The U.S. Sentencing Com-mission’s Attempts to Achieve Sentencing Reform in the Midst of Inter-Branch Power Struggles, 26J.L. & POL. 305, 346-50 (2011).


On the other side of the debate, some scholars have suggested that Booker im-proved the “quality, transparency, and rationality” in federal sentencing, and thusBooker is the “fix.”25 Indeed, the vast majority of federal district court judgesalso indicate that they prefer the current advisory Guidelines system to alternativeregimes. In a 2010 Sentencing Commission survey of district court judges, 75%of judges indicated that the current advisory Guidelines system “best achieves thepurposes of sentencing,” while only 3% preferred the mandatory Guidelines regimein place before Booker.26

This Article asks the question: What has been the result of reintroducing greaterjudicial discretion after Booker on inter-judge disparities? The primary empiricalwork on the impact of Booker on sentencing disparities is suggestive of increasesin inter-judge sentencing disparity. Using sentencing data from the District of Mas-sachusetts, Scott observes an increase in judge differences.27 While a first step, thestudy is limited to ten judges in the Boston courthouse. Therefore, a comprehensiveanalysis of disparities post Booker is essential to informing recent policy debates.28

25See Amy Baron-Evans & Kate Stith, Booker Rules, 160 U. PA. L. REV. 1631, 1633 (2012).26U.S. Sentencing Commission, RESULTS OF SURVEY OF UNITED STATES DISTRICT JUDGES

JANUARY 2010 THROUGH MARCH 2010 (June 2010) (Question 19, Table 19).27See Ryan W. Scott, Inter-judge Sentencing Disparity After Booker: A First Look, 63 STAN. L.

REV. 1, 4-5 (2010). Scott finds almost a doubling of the effect of judge on sentence length postBooker, resulting in an average difference of over two years between lenient and harsh judges forcases not subject to a mandatory minimum. Id. at 40-41. Scott also finds significant variation inthe rate of below-range sentencing among judges. Some judges sentenced below-range at the samerate prior to Booker (around 16%), while others increased their rate of sentencing below-range toas high as 53%. Id. The Transactional Records Access Clearinghouse (TRAC) recently compileda dataset of the sentencing records of over 800 federal judges from fiscal year 2007 to 2011. SeeSusan B. Long & David Burnham, TRAC Report: Examining Current Federal Sentencing Practices:A National Study of Differences Among Judges, 25 FED. SENTENCING REP. 6, 7 (2012); see alsoBig Sentencing Disparity Seen for Judges, N.Y. TIMES, March 5, 2012, at A23. Relying on the ran-dom assignment of cases to judges within district courthouses, the TRAC study found statisticallysignificant, unexplained differences in the typical sentences of judges in over half of the courthousesstudied. Id. at 15. The most recent Commission report (2012) also finds suggestive evidence thatvariation among judges within the same district, in particular the rates of non-government sponsoredbelow range sentences, has increased after Booker and Gall. United States Sentencing Commission,REPORT ON THE CONTINUING IMPACT OF UNITED STATES V. BOOKER ON FEDERAL SEN-TENCING (2012). The Commission concludes that “sentencing outcomes increasingly depend uponthe judge to whom the case is assigned.” Id. at 7. However, the Commission does not account forcaseload composition differences across judges within the same district, and analyzes all 94 districts,despite evidence by previous researchers that random assignment of cases is not universal. Thus, theCommission’s findings are only suggestive.

28Another strand of empirical research analyzes the impact of Booker on racial disparities insentencing. The United States Sentencing Commission has found evidence of large racial dis-parities in sentencing outcomes after Booker. See United States Sentencing Commission, DEMO-GRAPHIC DIFFERENCES IN FEDERAL SENTENCING PRACTICES: AN UPDATE OF THE BOOKERREPORT’S MULTIVARIATE REGRESSION ANALYSIS (2010) (providing evidence that demographicdifferences were significantly less when the Guidelines were binding, particularly during the PRO-TECT Act when appellate review of departures involved de novo review); United States SentencingCommission, FINAL REPORT ON THE IMPACT OF UNITED STATES V. BOOKER ON FEDERALSENTENCING, at 105-108 (2006). However, other scholars have found no significant change inracial disparities, at least in sentence length. See J.T. Ulmer et al., Racial Disparity in the Wakeof the Booker/Fanfan Decision: An Alternative Analysis to the USSC’s 2010 Report, 10 CRIM. &


This Article fills this gap by undertaking the first national, multi-district empir-ical analysis of inter-judge sentencing disparities in federal sentencing after Bookerby utilizing a new and comprehensive dataset constructed for this study. The Arti-cle proceeds in five parts. Part I provides a brief background of the legal landscape.Part II describes the framework, dataset and empirical methods. Matching threedata sources, I construct a dataset of over 600,000 criminal defendants linked tosentencing judge from 2000-2009.

Part III presents empirical results. Relying on the random assignment of casesto judges in district courthouses, I find evidence of significant increases in inter-judge disparities. A defendant who is randomly assigned a one standard deviation“harsher” judge in the district court received a 2.6 month longer prison sentencecompared to the average judge before Booker, but received a 5.3 month longer sen-tence following Kimbrough/Gall, a doubling of inter-judge disparities. Similarly,a defendant randomly assigned to a one standard deviation more “lenient” judgefaced a 4.8% chance of receiving a below range departure before Booker, but over6.6% chance after Kimbrough/Gall.

Part III.D undertakes an analysis of the sources of increases in inter-judge dis-parities. Many scholars have suggested that judges have different sentencing philoso-phies,29 which may be affected by the standard of appellate review.30 Sentencingpractices are correlated with judge demographic characteristics such as race,31 gen-

PUB. POL’Y 1077 (2011). Some scholars argue that judicial discretion may actually mitigate recentincreases in racial disparities. See Joshua Fischman & Max Schanzenbach, Racial Disparities underthe Federal Sentencing Guidelines: The Role of Judicial Discretion and Mandatory Minimums, 9 J.EMPIRICAL LEGAL STUD. 729 (2012) (arguing that recent increases in racial disparities are mainlydue to the increased relevance of mandatory minimums).

29See John S. Carroll et al., Sentencing Goals, Casual Attributions, Ideology, and Personality,52 J. PERSONALITY & SOC. PSYCHOL. 107 (1987) (arguing that judicial ideology is reflectedin how a judge thinks about the causes of crime and the goals of sentencing); Shari S. Diamond &Hans Zeisel, Sentencing Councils: A Study of Sentence Disparity and Its Reduction, 43 U. CHI. L.REV. 109, 114 (1975) (“it is reasonable to infer that the judges’ differing sentencing philosophiesare a primary cause of the disparity"); see also, Paul Hofer, Kevin Blackwell, & R. Barry Ruback,The Effect of the Federal Sentencing Guidelines on Inter-Judge Sentencing Disparity, 90 J. CRIM.L. & CRIMINOLOGY 239 (1999) (claiming that there are differences between how liberals andconservatives view the goals of sentencing which can drive different sentencing practices).

30Joshua Fischman & Max Schanzenbach, Do Standards of Review Matter? The Case of FederalCriminal Sentencing, 40 J. LEG. STUD. 405 (2011). The authors find that Democratic appointeesare more lenient than Republican appointees and differences in sentencing practices increase whenappellate review is more deferential, suggesting that judges are constrained by the fear of reversal.The authors also find evidence that pre-Guidelines appointed judges are more likely to depart fromthe Guidelines than post Guidelines appointees.

31See Thomas Uhlman, RACIAL JUSTICE: BLACK JUDGES AND DEFENDANTS IN AN UR-BAN TRIAL COURT (1979) (claiming that black and white judges sentenced black defendants moreharshly compared to white defendants); Susan Welch et al., Do Black Judges Make a Difference?,32 AM. J. POL. SCI. 126 (1988) (finding that black judges do not differ much in their incarcerationdecisions from white judges based on one city’s criminal cases).


der,32 and political affiliation.33 In particular, the inevitable shift in the compositionof the federal district courts may have profound consequences on unwarranted dis-parities as judges who have no experience sentencing under a presumptive Guide-lines regime take the federal bench.34 Federal defense lawyer James Felman pre-dicted that following the introduction of the advisory Guidelines, “unwarranted dis-parity in the near term would be considerably less than that which existed prior to1987,” but “there will be a minority of judges who will generate unwarranted dis-parity, and this number seems likely to increase as the years go by and the bench isfilled with individuals who have no history with binding guidelines.”35

I find that female judges and Democratic appointed judges issue shorter sen-tences and are more likely to depart downwards from the Guidelines than theirmale and Republican appointed peers, respectively. Also striking are the differen-tial sentencing practices of post Booker and pre-Booker judicial appointees. Judgeswho have no prior experience sentencing under the mandatory Guidelines regimeare more likely to depart from the Guidelines recommended range than their pre-Booker counterparts, potentially suggesting that newer judges are less anchored tothe Guidelines.

In addition to analyzing the impact of philosophy on sentencing practices ofindividual judges, this paper also contributes to the literature on geographical vari-ations in sentencing patterns, which has found that court cultures can affect sen-tencing through both exercise of judicial discretion and also regional differencesin policies among prosecutors.36 In Part III.E, I present evidence of substantial

32See, e.g., Darrell Steffensmeier & Chris Herbert, Women and Men Policy Makers: Does theJudge’s Gender Affect the Sentencing of Criminal Defendants?, 77 SOCIAL FORCES 1163 (1999)(female judges sentence defendants for longer terms, are more likely to incarcerate minorities, andless likely to incarcerate women, in Pennsylvania criminal cases); Max Schanzenbach, Racial andSex Disparities in Sentencing: The Effect of District-Level Judicial Demographics, 34 J. LEG.STUD. 57 (2005) (some evidence that minority and female judges sentence differently using districtlevel variation in judicial characteristics).

33For instance, Max M. Schanzenbach & Emerson H. Tiller, Strategic Judging Under the UnitedStates Sentencing Guidelines: Positive Theory and Evidence, 23 J.L. ECON. & ORG. 24 (2007) ex-plore the impact of ideology on federal criminal sentencing decisions from 1992 through 2001. Theyfind that sentences for serious crimes in districts comprised of more Democrat appointed judges arelower than sentences in districts with more Republican appointed judges. The alignment of the ide-ology of the reviewing court also increased departures from the Guidelines. More recent work inMax M. Schanzenbach & Emerson H. Tiller, Reviewing the Sentencing Guidelines: Judicial Pol-itics, Empirical Evidence, and Reform, 75 U. CHI. L. REV. 715 (2008) reveals that Republicanappointed judges give longer sentences for the same crime compared to their Democratic appointedcounterparts. Moreover, Democratic-appointed judges are more likely to depart downwards fromthe Guidelines when the reviewing circuit court is majority Democratic appointed.

34Until now, prior studies have been unable to identify the impact of post Booker appointedjudges on inter-judge sentencing disparities. The Scott study, which only looks at the Boston court-house, is unable to take into account changes in judicial composition because the Boston courthousedid not experience turnover during the years in his study. Recent work suggest that racial disparitiesin sentencing are greater among judges appointed after Booker. See Crystal S. Yang, Free At Last?Judicial Discretion and Racial Disparities in Federal Sentencing, Harvard John M. Olin Center forLaw, Economics, and Business Fellows’ Discussion Paper Series No. 47 (2012).

35See James Felman, How Should Congress Respond if the Supreme Court Strikes Down theFederal Sentencing Guidelines?, 17 FED. SENTENCING REP. 97, 98-99 (2004).

36See, e.g., Paul L. Sutton, Department of Justice, Federal Sentencing Patterns: A Study of Geo-


inter-district differences in sentencing outcomes. Significant differences exist insentence length, rates of below range departures, rates of mandatory minimums,and rates of substantial assistance motions, with inter-district differences expand-ing after Booker.

In Part III.F, I present some evidence on the contribution of prosecutorial deci-sions on inter-judge disparities. Undoubtedly, a defendant’s sentence is determinedby the discretionary actions of multiple actors in the criminal justice process, cul-minating in sentencing. Therefore, any study of inter-judge sentencing disparitiesis only a partial portrayal of the disparities that can arise in the criminal justicesystem. Previous scholars rightly suggested that arrest, charge, and plea bargainingdecisions made earlier in the process are all ripe avenues for unwarranted bias.37

In particular, I analyze the impact of mandatory minimums on inter-judge dis-parities. If mandatory minimums were charged prior to judge assignment, onewould in fact expect the application of mandatory minimums to reduce inter-judgedisparities.38 However, prosecutorial discretion can lead to disparate application ofmandatory minimums across judges, potentially through the use of superseding in-dictments filed after judge assignment to the case.39 Indeed, I find evidence that theapplication of a mandatory minimum is a large contributor of inter-judge and inter-district disparities, such that measures of disparity are reduced by almost a factor oftwo when I exclude cases in which mandatory minimums are charged. The resultssuggest substantial unequal application of mandatory minimums to similar caseswithin district courthouses, and different mandatory minimum policies by prose-cutors across district courts. There are also substantial differences in the rates of

graphical Variations (1978); William M. Rhodes & Catherine Conly, Department of Justice FederalJustice Research Program, Analysis of Federal Sentencing (1981); Charles D. Weisselberg & Ter-rence Dunworth, Inter-District Variation under the Guidelines: The Trees May Be More SignificantThan the Forest, 6 FED. SENTENCING REP. 25, 26-27 (1993) (finding that the Guidelines do notimpact all cases and all districts equally and that the Guidelines mean different things in differentcontexts).

37See, e.g., Franklin E. Zimring, Making the Punishment Fit the Crime, 6 HASTINGS CENTERREP. 13, 13-14 (1976) (arguing that there is “multiple discretion” in the criminal justice systemfrom the legislature, prosecutor, judge and parole board); Ilene H. Nagel & Steven J. Schulhofer,A Tale of Three Cities: An Empirical Study of Charging and Bargaining Practices Under the Sen-tencing Guidelines, 66 S. CAL. L. REV. 501, 502 (1992) (noting that “both Congress and the U.S.Sentencing Commission were well aware that plea bargaining posed a potential threat to the successof guidelines sentencing”). As a result, the Guidelines system has been attacked by many for itsrigidity and for shifting power to prosecutors in their charge and plea bargaining decisions. See,e.g., Daniel J. Freed, Federal Sentencing in the Wake of the Guidelines: Unacceptable Limits on theDiscretion of Sentencers, 101 YALE L.J. 1681, 1719-20, 1725-27 (1992); Kate Stith, The Arc of thePendulum: Judges, Prosecutors, and the Exercise of Discretion, 117 YALE L.J. 1420, 1430 (2008);Albert W. Alschuler, Sentencing Reform and Prosecutorial Power: A Critique of Recent Proposalsfor “Fixed” and “Presumptive” Sentencing, 126 U. PA. L. REV. 550 (1978); Kate Stith & Jose A.Cabranes, FEAR OF JUDGING: SENTENCING GUIDELINES IN THE FEDERAL COURTS (1998).

38See infra Part III.F.39For instance, prosecutors often bring superseding indictments under various statutes, such as

20 U.S.C. §851 and 18 U.S.C. §924(c). A literature documents the large degree of prosecutorialdiscretion in bringing superseding indictments and potential due process concerns. See, e.g., DonaldC. Smaltz, Due Process Limitations on Prosecutorial Discretion in Re-Charging Defendants: Pearceto Blackledge to Bordenkircher, 36 Wash. & Lee L. Rev. 347, 353 (1979) (“Absent an adequatejustification for the superseding or additional charges, vindictiveness will be presumed.”).


substantial assistance motions filed by prosecutors across judges and district courts.Such results indicate that prosecutorial charging is a measurable contributor to dis-parities in sentencing.

In Part IV, I describe recent proposals to reform federal sentencing and applythe empirical findings in this paper to shed light on the soundness of the propos-als. Given the finding that a substantial portion of inter-judge disparities and re-gional disparity is attributable to the application of mandatory minimums, any pro-posal that contemplates shifting power to prosecutors will likely exacerbate thepresence of disparities. Indeed, many judges and scholars have suggested thatmandatory minimums are “fundamentally inconsistent with the sentencing guide-lines system.”40 Instead, I argue that strengthened appellate review and eliminationof mandatory minimums are potential steps in the direction of reducing unwarranteddisparities in sentencing. Part V concludes.

I BRIEF LEGAL BACKGROUND OF FEDERAL SENTENCING

A. Adoption of the Federal Sentencing Guidelines

In the early twentieth century, criminal justice was premised on the notion of re-habilitation.41 This goal of rehabilitation manifested itself in the form of indeter-minate sentencing, which allowed prison sentences and probation to be tailored toeach offender’s progress toward reform. As a result, judges and parole boards pos-sessed substantial discretion in their sentencing determinations.42 In this regimeof “free at last” sentencing,43 federal judges had essentially unlimited authority inimposing sentences, restrained only by statutorily prescribed minimum and maxi-

40See Sessions, supra note 24, at 329 (citing Senator Edward M. Kennedy, Sentencing Reform-An Evolutionary Process, 3 FED. SENTENCING REP. 271, 272 (1991) (“Mandatory minimumsentencing statutes have . . . hampered the guideline system and are becoming an increasinglyserious obstacle to its success. . . . Mandatory minimums inevitably lead to sentencing disparitybecause defendants with different degrees of guilt and different criminal records receive the samesentence.”)); Stephen Breyer, Federal Sentencing Guidelines Revisited, 11 FED. SENTENCING REP.180, 184-85 (1999) (“. . . . Congress, in simultaneously requiring Guideline sentencing andmandatory minimum sentences, is riding two different horses. And those two horses, in terms ofcoherence, fairness, and effectiveness, are traveling in opposite directions. . . . . [Congress needsto] abolish mandatory minimums altogether.”).

41See David Rothman, Sentencing Reform in Historical Perspective, 29 CRIME & DELIN-QUENCY 631, 637-41 (1983) (reformers “pursue[d] rehabilitation, which meant treating the criminalnot the crime, calculating the sentence to fit the individual needs and characteristics of the offender”);see also Frank O. Bowman, III, The Quality of Mercy Must Be Restrained, and Other Lessons inLearning to Love the Federal Sentencing Guidelines, 1996 WIS. L. REV. 679, 680-89 (describingthe rehabilitative sentencing model).

42See Rothman, supra note 41, at 638 (“the judge would make his decision (probation or such aminimum-maximum term); eventually the prison classification committee experts would make theirdecision (this program or that program), and the parole board experts would make theirs (release atthe minimum, or later)”).

43A term coined by Judge Nancy Gertner to describe the state of indeterminate sentencing priorto 1984. See United States v. Mueffelman, 400 F. Supp. 2d 368, 372 (D. Mass 2005); United Statesv. Jaber, 362 F. Supp. 2d 365, 370 (D. Mass. 2005).


mum sentences.44 Lack of appellate review of sentences meant that judges faced nomeaningful check to ensure uniformity and consistency in sentencing.45

By the 1970s, faith in the rehabilitative model of sentencing declined due to aconfluence of changing social norms, escalating public anxiety over rising crimerates, and public skepticism of the ability to rehabilitate criminal offenders.46 Thelegal community and public expressed alarm at the widespread disparities in crim-inal sentencing. Some argued that judges and parole boards endangered publicsafety with lenient sentencing of criminals.47 Others were distressed by inequitableand arbitrary treatment within the criminal justice system as studies showed thatsimilar offenders were often punished very differently. A 1977 study of 47 Virginiastate district court judges revealed that while judges generally agreed on the verdictin legal cases, they applied radically different sentences.48 A Federal Judicial Cen-ter Second Circuit Study found large inter-judge differences in the sentences im-posed based on identical presentence reports of defendants.49 The same defendantwas sentenced to three years by one judge, and twenty years by another.50 Someconcluded that this disparate treatment of defendants by judges produced racial in-equities in sentencing. The American Friends Service Committee claimed that de-creasing discretion among judges and parole boards was the only way to eliminateracial discrimination and sentencing disparities in the criminal justice system.51

Other studies identified large inter-court differences. A 1988 study of federalcourts found that white collar offenders who committed similar offenses receivedvery different sentences depending on the court in which they were sentenced,52

with one study observing that “ a defendant sentenced by a Southern judge was

44See United States Sentencing Commission, THE FEDERAL SENTENCING GUIDELINES: AREPORT ON THE OPERATION OF THE GUIDELINES SYSTEM AND SHORT-TERM IMPACTS ONDISPARITY IN SENTENCING, USE OF INCARCERATION, AND PROSECUTORIAL DISCRETIONAND PLEA BARGAINING, Vol. I at 9 (December 1991) (judges “decided the various goals of sen-tencing, the relevant aggravating and mitigating circumstances, and the way in which these factorswould be combined in determining a specific sentence”); see Koon v. United States, 518 U.S. 81, 96(1996) (“Before the Guidelines system, a federal criminal sentence within statutory limits was, forall practical purposes, not reviewable on appeal.”)

45See, e.g., Dorszynski v. United States, 418 U.S. 424, 431 (1974) (stating the general propositionthat appellate review ends if a sentence is within the limitations set forth in the statute).

46See, e.g., Francis A. Allen, THE DECLINE OF THE REHABILITATIVE IDEAL: PENAL POLICYAND SOCIAL PURPOSE, 25-30 (1981); see also Bowman, Debacle, supra note 22, at 374 (attributingdemand for social controls to rising crime rates and social upheaval).

47Michael Tonry, Obsolescence and Immanence in Penal Theory and Policy, 105 COLUM. L.REV. 1233, 1247 (2005) (conservatives criticized indeterminate sentencing for being uncertain andlenient and increasing crime rates).

48William Austin & Thomas A. Williams, III, A Survey of Judges’ Responses to Simulated LegalCases: Research Note on Sentencing Disparity, 69 J. CRIM. L. & CRIMINOLOGY 306 (1977).

49See Partridge & Eldridge, supra note 2, at 36.50Id.51See Am. Friends Serv. Comm., Struggle for Justice: A Report on Crime and Punishment in

America 130 (1971) (claiming that discretion allowed judges and parole boards to control minoritygroups); see also Bowman, Quality of Mercy, supra note 41, at 686-88 (critics arguing discretionproduced unjustifiable disparities open to conscious or unconscious racial and other biases, demand-ing “truth in sentencing”).

52Wheeler et al., SITTING IN JUDGMENT: THE SENTENCING OF WHITE-COLLAR CRIMINALS(1988).


likely to serve six months more than average, while a defendant sentenced in CentralCalifornia was likely to serve twelve months less.”53

These large disparities in sentencing prompted calls for sentencing reform. Cham-pioning the call for reform, federal district judge Marvin Frankel expressed graveconcern over the indeterminate and individualized sentencing regime of the pe-riod.54 Judge Frankel called for the creation of an independent sentencing commis-sion that would replace judicial and parole board discretion.55

In response, Congress created the United States Sentencing Commission toadopt and administer the Sentencing Guidelines, aimed at eliminating unwarrantedsentencing disparities “among defendants with similar records who have been foundguilty of similar criminal conduct.”56 Part of the SRA of 1984, the Guidelines wereapplied to all federal offenses committed after November 1, 1987. The SRA hasbeen viewed by some as creating a regime that preserves judicial discretion,57 whileothers have viewed the SRA has substantially increasing the severity of punishmentand dramatically reducing the discretion afforded to judges to consider the particu-lar circumstances of each offender.58

Notwithstanding disagreement about the degree to which sentencing reformchanged the legal landscape, the new SRA introduced a shift from a regime of vir-tually unfettered judicial discretion to more restricted discretion within a system ofdeterminate sentencing.59 By requiring judges to sentence within the recommendedGuidelines range unless the court found aggravating or mitigating circumstances,60

the Guidelines were treated as presumptively mandatory, although the particularstandards for departure were ambiguous.61 Later in Koon v. United States, theSupreme Court clarified that a district court judge’s decision to depart from theGuidelines range would be subject to an abuse of discretion standard of appellatereview.62

53See Justice Stephen Breyer, Federal Sentencing Guidelines Revisited, 11 FED. SENTENCINGREP. 180 (1988).

54Frankel, supra note 3, at 5. Frankel also argued that individualized sentencing was “out ofhand,” and criticized the state of indeterminate sentencing. Id. at 10, 26-49; see also Marvin E.Frankel, Lawlessness in Sentencing, 41 U. CIN. L. REV. 1 (1972).

55Frankel, supra note 3.5628 U.S.C. §991(b)(1)(B).57Marc L. Miller & Ronald F. Wright, Your Cheatin’ Heart(land): The Long Search for Admin-

istrative Sentencing Justice, 2 BUFF. CRIM. L. REV. 723 (1999).58See Stith & Koh, supra note 5, at 284-85 (“It should come as a surprise to no one that in those

areas where the statute is ambiguous or otherwise deliberately leaves important policy issues to theCommission, the Commission has generally chosen to increase the rigidity and complexity of itsguidelines. It is no accident that judges have found it difficult to depart from the guidelines; this isprecisely what Congress intended.” ).

59In addition to creating the Guidelines, the SRA also abolished federal parole and institutedsupervised release in its place. Supervised release is meant “to assist individuals in their transitionto community life,” and “fulfills rehabilitative ends, distinct from those served by incarceration.”United States v. Johnson, 529 U.S. 53, 59 (2000). The term of supervised release is imposed alongwith a prison term at the time of sentencing.

60H.R. REP. NO. 1017, 98th Congress, 2nd Session, at 93-94 (1984).61See Miller & Wright, supra note 57, at 730 (The Commission allowed judges to depart from

the Guidelines when the case fell within the “heartland,” but the concept was left highly general.)62518 U.S. 81, 99 (1996). The Court in Koon stated that Congress “did not intend, by establish-


Under the Guidelines, federal district court judges assign each federal crime toone of 43 offense levels, and assign each federal defendant to one of six criminalhistory categories. The more serious the offense and the greater the harm associatedwith the offense, the higher the base offense level assigned under Chapter Two ofthe Guidelines.63 For example, trespass offenses are assigned a base offense levelof four,64 while kidnapping is assigned a base offense level of 32.65 From a baseoffense level, the final offense level is calculated by adjusting for applicable offensecharacteristics and adjustments. Relevant adjustments include the amount of lossinvolved in the offense, use of a firearm, the age or condition of the victim, etc.66

Chapter Three of the Guidelines allows for further adjustments based on aggravat-ing or mitigating factors, such as a defendant’s acceptance of responsibility.67

The criminal history category reflects the frequency and severity of a defen-dant’s prior criminal convictions. To determine a defendant’s criminal history cate-gory, a judge adds points for prior sentences in the federal system, 50 state systems,all territories and foreign or military courts.68 For example, three points are addedfor each prior sentence of imprisonment exceeding one year and one month, andtwo points are added for each prior sentence of imprisonment of at least 60 daysand less than one year and one month.69 Two points are also added if the defendantcommitted the instant offense under any criminal justice sentence.70

The intersection of the final offense level and criminal history category yieldsa fairly narrow Guidelines recommended sentencing range, where the top of therange exceeds the bottom by the greater of either six months or 25%. If a judgedetermines that there are aggravating or mitigating circumstances that warrant adeparture from the Guidelines, she would have to justify her reasons for departureto the appellate court,71 but in general the Guidelines were treated as sufficiently

ing limited appellate review, to vest in appellate courts wide-ranging authority over district courtsentencing decisions.” Id. at 97.

63United States Sentencing Commission, Guidelines Manual, Chapter Two.64Id. §2B2.3.65Id. §2A4.1.66Id.67For instance, the Guidelines allow for a decrease in base offense level for a defendant’s accep-

tance of responsibility under §3E1.1 or for minimal participation in the offense under §3B1.2. Baseoffense level is increased for defendants who obstruct or impede the administration of justice under§3C1.1.

68United States Sentencing Commission, Guidelines Manual, §4.1 (“A defendant with a recordof prior criminal behavior is more culpable than a first offender and thus deserving of greater pun-ishment. Greater deterrence of criminal conduct dictates that a clear message be sent to society thatrepeated criminal behavior will aggravate the need for punishment with each recurrence. To protectthe public from further crimes of the particular defendant, the likelihood of recidivism and the fu-ture criminal behavior must be considered. Repeated criminal behavior is an indicator of a limitedlikelihood of successful rehabilitation.”).

69Id. §4.1.70Id. §4.1.7118 U.S.C. §3553(b) (“the court shall impose a sentence of the kind, and within the range, re-

ferred to in subsection (a)(4) unless the court finds that there exists an aggravating or mitigatingcircumstance of a kind, or to a degree, not adequately taken into consideration by the SentencingCommission in formulating the guidelines that should result in a sentence different from that de-scribed").


mandatory prior to Booker.72 After the imposition of a sentence, the governmentis permitted to appeal any sentence resulting in a departure below the Guidelinesrange, and the defendant can appeal an upward departure.73

There are numerous other ways in which Congress has attempted to limit un-warranted disparities in sentencing. Beginning in 1984, and subsequently 1986and 1988, Congress enacted a series of mandatory minimum statutes directed atdrug and firearms offenses.74 Mandatory minimums were also applied to recidi-vist offenders, through the Armed Career Criminal Act,75 enhancements for careeroffenders,76 and enhancements for repeat and dangerous sex offenders.77

In 2003, Congress passed the PROTECT Act to curtail judicial departures dueto a concern that the standard for appellate review of departures had led to un-desirably high rates of downward departures for child sex offenses.78 Under theFeeney Amendment to the PROTECT Act, judicial departures were only allowedfor certain reasons outlined in the Guidelines Manual.79 Additionally, the FeeneyAmendment to the PROTECT Act replaced the prior abuse of discretion standard ofreview for downward departures with de novo review by overturning the SupremeCourt’s holding in Koon.

B. Challenges to the Mandatory Guidelines Regime

The initial challenge to the federal sentencing regime began with the “watershed”ruling in Apprendi v. New Jersey.80 In Apprendi, the Supreme Court found a NewJersey hate crime statute unconstitutional because it authorized judges to imposehigher sentences based on facts other than those submitted to a jury, and provedbeyond a reasonable doubt.81

72The Court in Booker noted that “[t]he Guidelines as written ... are not advisory; they aremandatory and binding on all judges” and therefore “have the force and effect of laws.” Booker, 543U.S. at 233.

7318 U.S.C. §3742 (a)-(b).74Anti-Drug Abuse Act of 1986 (Pub. L. 99-570, 100 Stat. 3207); Pub. L. No. 98-473, §1005(a),

98 Stat. 1837, 2138-39 (1986).75Guidelines Manual §4B1.2; 18 U.S.C. §924(e). The Armed Career Criminal Act (ACCA)

imposes a minimum 15-year term of imprisonment for defendants convicted of unlawful possessionof a firearm under 18 U.S.C. §922(g), with three prior state or federal convictions for violent feloniesor serious drug offenses.

76§4B1.1; 28 U.S.C. §944(h) (mandating that the Commission impose imprisonment “at or nearthe maximum term authorized” for defined “career” offenders).

77§4B1.5; 18 U.S.C. §§2247, 2426.78Pub. L. 108-21, 117 Stat. 650, S. 151, 2003.79For certain offenses, such as child abduction and child sex offenses, the PROTECT Act

amended 18 U.S.C. §3553(b) to only allow the sentencing court to depart downwards if there aremitigating circumstances of a kind or to a agree that has been “affirmatively and specifically identi-fied” as permissible grounds for downward departure. The PROTECT Act also amended the Guide-lines Manual §5K2.0 to state that the “the grounds enumerated in this Part K of Chapter Five arethe sole grounds that have been affirmatively and specifically identified as a permissible ground ofdeparture.”

80530 U.S. 466, 425 (2000) (O’Connor, J., dissenting) (calling the Apprendi decision a“watershedchange in constitutional law”).

81Id. at 468-69, 490.


These principles were subsequently applied to the constitutionality of manda-tory sentencing guidelines, first questioned in reference to the Washington StateSentencing Guidelines. In Blakely v. Washington, the Supreme Court held thatthe Sixth Amendment right to a jury trial prohibited judges from increasing a de-fendant’s sentence beyond the statutory maximum based on facts other than thosedecided by the jury beyond a reasonable doubt.82 As a result, Washington’s manda-tory sentencing guidelines were struck down. While the majority opinion in Blakelyemphasized that the decision was not passing judgment on the constitutionality ofthe Federal Sentencing Guidelines,83 the parallels were apparent and shortly after,the reasoning of Blakely was applied to the Guidelines.

In United States v. Booker, the mandatory Federal Sentencing Guidelines werealso found to be unconstitutional under the Sixth Amendment by mandating judi-cial fact-finding in determining sentencing ranges.84 The Booker ruling, however,did not apply to Congressionally-enacted mandatory minimum sentences.85 Ratherthan invalidate the Guidelines wholly, or prescribe an enhanced role for jury fact-finding, the Court held in a separate remedial decision led by Justice Breyer, thatthe remedy for the Sixth Amendment violation was to declare the Guidelines nolonger mandatory but “effectively advisory.”86 The Court explained that “districtcourts, while not bound to apply the Guidelines, must consult those Guidelines andtake them into account when sentencing.”87

In the immediate aftermath of Booker, district courts took varied approaches inapplying Booker.88 Some courts sentenced with minimal consideration of the appli-cable Guidelines range, while others treated the Guidelines as a dominant factor.89

Circuit courts later reached a consensus that sentencing must begin with the cal-culation of the applicable Guidelines range.90 Today, after a sentencing judge has

82542 U.S. 296 (2004).83Id. at 305, n.9 (“The Federal Guidelines are not before us, and we express no opinion on

them”).84543 U.S. 220, 226-27, 243-44 (2005) (Stevens, J., writing for the Court).85See United States Sentencing Commission, REPORT TO CONGRESS: MANDATORY MINI-

MUM PENALTIES IN THE FEDERAL CRIMINAL JUSTICE SYSTEM, at 37 (Oct. 2011) (discussionof the compatibility of mandatory minimums and the Guidelines after Booker).

86Booker, 543 U.S. at 245 (Breyer, J., writing for the Court). Similarly, the provisions on super-vised release also became advisory, although the USSC states that the majority of courts continue toimpose at least the minimum terms in §5D1.2.

87Booker, 543 U.S. at 264.88See Gilles R. Bissonnette, “Consulting” the Federal Sentencing Guidelines After Booker, 53

UCLA L. REV. 1497, 1521-22 (2006) (arguing that Booker left open how much sentencing judgescould deviate from the Guidelines).

89See id. at 1522-32 (claiming that district courts have taken two approaches in applying Booker,the “substantial-weight” approach and the “consultative” approach). For an example of the two typeof approaches taken by district courts, see, e.g. United States v. Ranum, 353 F. Supp. 2d 984, 987(E.D. Wis. 2005) (noting that “Booker is not ... an invitation to do business as usual,” but that courtsshould consider all the factors in §3553(a)); cf. United States v. Wilson, 350 F. Supp. 2d 910, 925(D. Utah 2005) (suggesting courts give “heavy weight” to the Guidelines after Booker).

90See, e.g., United States v. Howard, 454 F.3d 700, 703 (7th Cir. 2006) (court noting that afterBooker, “the district court must first calculate the proper Guidelines range and then, by referenceto the factors specified in 18 U.S.C. §3553(a), select an appropriate sentence”); United States v.Crosby, 397 F.3d 103, 113-114 (2d Cir. 2005) (“consideration of the Guidelines will normally


calculated the applicable Guidelines range, he or she must consider seven factorsbefore imposition of punishment: (1) the nature and circumstances of the offenseand the history and characteristics of the defendant, (2) the need for the sentenceimposed, (3) the kinds of sentences available, (4) the kinds of sentence and thesentencing range established, (5) any pertinent policy statement issued by the Sen-tencing Commission, (6) the need to avoid unwarranted sentence disparities amongdefendants with similar records who have been found guilty of similar conduct, and(7) the need to provide restitution to any victims of the offense.91 After consid-eration of all the factors, the sentencing judge is instructed to “impose a sentencesufficient, but not greater than necessary, to comply with” the basic goals of sen-tencing.92

Subsequent Supreme Court decisions furthered weakened the effect of the Guide-lines on criminal sentencing. While the Court in Booker declared that departuresfrom the Guidelines should be reviewed under a “reasonableness” standard, it didnot clarify the meaning of this standard until a sequence of decisions in 2007. InRita v. United States, the Court held that a sentence within the Guidelines recom-mended range could be presumed “reasonable” because a “judge who imposes asentence within the range recommended by the Guidelines thus makes a decisionthat is fully consistent with the Commission’s judgment in general.”93 In Gall v.United States, the Court further held that federal appeals courts could not presumethat a sentence outside the range recommended by the Guidelines was unreason-able, reducing the degree of appellate review.94 The Court in Gall concluded thatin reviewing a sentence outside the Guidelines range, an appellate court could con-sider the extent of deviation from the Guidelines, but must give “due deference tothe district court’s decision that the §3553(a) factors, on a whole, justify the ex-tent of the variance.”95 In the aftermath of Gall, appellate courts could only reviewsentencing decisions under the more deferential abuse of discretion standard. Con-current with the Gall decision, the Supreme Court confirmed the holding in Bookeras applied to cases involving possession, distribution and manufacture of crack co-caine.96 The Court in Kimbrough held that federal district court judges have thediscretion to impose sentences outside the recommended Guidelines range due topolicy disagreements with the Sentencing Commission.97

require determination of the applicable Guidelines range, or at least identification of the arguablyapplicable ranges ... it would be a mistake to think that, after Booker/Fanfan, district judges mayreturn to the sentencing regime that existed before 1987 and exercise unfettered discretion to selectany sentence within the applicable statutory maximum and minimum”).

9118 U.S.C. §3553(a).92Id.93551 U.S. 338, 347-50 (2007) (holding that “a court of appeals may apply a presumption of

reasonableness to a district court sentence that reflects a proper application of the Sentencing Guide-lines”).

94552 U.S. 38 (2007).95Id. at 51 (arguing the an appellate court’s disagreement with the appropriateness of a sentence

is “insufficient to justify reversal”).96Kimbrough v. United States, 552 U.S. 85 (2007).97Id. (granting sentencing judges explicit permission to deviate from the Guidelines based on

disagreement with the disparate treatment of crack and powder cocaine offenses - the so called“100-to-1 ratio”).


II FRAMEWORK, DATA, AND METHODS

A. Judicial Behavior in Criminal Sentencing

While judges have an obligation to “follow the law,”98 they have additional mo-tivations that affect their decision-making.99 Scholars have suggested that judgescare about a variety of factors such as public recognition, leisure, and reputation.100

In addition, judges have preferences for sentencing according to their personaltastes.101 In the context of criminal sentencing, a judge may prefer to sentence adefendant based on personal, political or ideological goals, rather than according tothe mandated Guidelines sentence.102

How might the sentencing regime affect judicial behavior? Given individualjudge-specific preferences for sentencing, one would likely observe large inter-judge disparities if judges were left unconstrained in the exercise of discretion.Consistent with this prediction, scholars have suggested that the large variancesin federal sentences prior to the adoption of the Guidelines were likely due to dif-fering judicial attitudes regarding rehabilitation and deterrence.103 At the other endof the spectrum, judges who are restricted in the exercise of discretion would beunable to fully sentence according to their preferences. The type of sentencing

98See, e.g., Lewis A. Kornhauser, Judicial Organization and Administration, 7 Encyclopedia ofLaw and Economics 27 (2000)

99The economic analysis of judicial behavior builds on work by Judge Richard A. Posner. See,e.g. Hon. Richard A. Posner, Judicial Behavior and Performance: An Economic Approach, 32 FLA.ST. U. L. REV. 1259 (2005); see also Hon. Richard A. Posner, HOW JUDGES THINK (2008);Lee Epstein, William M. Landes & Richard A. Posner, THE BEHAVIOR OF FEDERAL JUDGES: ATHEORETICAL AND EMPIRICAL STUDY OF RATIONAL CHOICE (2013). See Nicola Gennaioli &Andrei Shleifer, Judicial Fact Discretion, 37 J. LEG. STUD. 1 (2008) for a theoretical economicmodel of judicial discretion in fact determination.

100Federal district judges occupy a unique position because most district judge appointments areterminal, thus “insulat[ing[ the judges from the normal incentives and constraints that determine thebehavior of rational actors, except for the relative handful of judges who are ambitious for promotionto the court of appeals.” Posner, Judicial Behavior and Performance: An Economic Approach, supranote 99, at 1260, 1269 (noting that a judge likely cares more about leisure and public recognitionrelative to income, compared to average practicing attorneys).

101Id. at 1269-70 (“deciding a particular case in a particular way might increase the judge’s utilityjust by the satisfaction that doing a good job produces ... [or] by advancing a political or ideologicalgoal”).

102Indeed, federal district court judges have expressed a great degree of dissatisfaction with theGuidelines. In a 2010 survey of federal district judges, 65% of judges indicated that they thoughtthe departure policy statements in the Guidelines Manual were too restrictive, suggesting that manyjudges would prefer to deviate from the Guidelines. U.S. Sentencing Commission, RESULTS OFSURVEY OF UNITED STATES DISTRICT JUDGES JANUARY 2010 THROUGH MARCH 2010 (June2010), (Question 14, Table 14).

103See Posner, Judicial Behavior and Performance: An Economic Approach, supra note 99, at1270 (inferring from the “extraordinary variance” in federal sentences prior to the promulgation ofthe Guidelines that differences in sentence lengths were due to judicial attitudes on responsibilityand deterrence); see also Brian Forst & Charles Wellford, Punishment and Sentencing: DevelopingSentencing Guidelines Empirically From Principles of Punishment, 33 RUTGERS L. REV. 799, 801-804 (1981) (documenting disagreement between judges regarding five goals of sentencing: generaldeterrence, special deterrence, incapacitation, rehabilitation, and just deserts).


regime can place a constraint on judicial discretion. For instance, the adoption ofdeterminate sentencing under the Guidelines introduced a mechanism by which toconstrain judges, likely explaining studies finding reduced inter-judge sentencingdisparity after the promulgation of the Guidelines.104

In addition to a restriction on free sentencing imposed by the sentencing regime,another constraint comes from the prospect of appellate review. Judges who departfrom the Guidelines incur economic and social costs from deviating. A high rever-sal rate is not only administratively burdensome, but also potentially harms a trialjudge’s prospects for promotion to the appeals court.105 Indeed, under the manda-tory sentencing regime, departures from the Guidelines were relatively rare.106 Af-ter the Feeney Amendment of the PROTECT Act, which subjected district courtjudges to de novo review for departures from the Guidelines, departures were re-duced even further, suggesting that judges respond to changes in appellate scrutiny.107

Given the countervailing forces of (1) judge sentencing preferences versus (2)costs of exercising discretion, what is the theoretical prediction of the impact ofBooker, Rita, Gall, and Kimbrough on inter-judge disparities? Following Booker,the total cost of exercising discretion fell substantially for judges who wanted todepart from the Guidelines sentence as the Guidelines were rendered advisory. Thismajor shift in sentencing may have been accompanied by increases in inter-judgedisparity.

However, to the extent the the relative cost associated with de novo appellatereview was still binding, judges may have been hesitant to alter their practices.Indeed, not until Rita, Gall, and Kimbrough did the Court reduce the level of appel-late review from de novo to substantial abuse of discretion, intuitively lowering theprobability of appellate reversal.108 Thus, one might expect larger increases in inter-judge disparities following Rita, Gall, and Kimbrough. Nevertheless, given the Ritapresumption of reasonableness attached to within range sentences, the Guidelinesstill provide a safe harbor from appellate scrutiny.

There are other reasons why judicial behavior and inter-judge disparities mayhave not changed much after Booker and its progeny. First, judges may becomeacculturated to the Guidelines if they have had substantial experience sentencing

104See supra notes 9-14.105See Posner, Judicial Behavior and Performance: An Economic Approach, supra note 99, at

1270-71; see also Stephen J. Choi, Mitu Gulati, & Eric A. Posner, What do Federal District JudgesWant?: An Analysis of Publications, Citations, and Reversals, University of Chicago John M. OlinLaw & Economics Working Paper Series Paper No. 508, at 3-4 (2009) (judges care about min-imizing workload and maximizing reputation by avoiding appellate reversal); Evan H. Caminker,Precedent and Prediction: The Forward-Looking Aspects of Inferior Court Decisionmaking, 73TEX. L. REV. 1, 77-78 (1994) (describing anecdotal evidence that lower court judges dislike beingreversed on appeal due to professional reputation, advancement, judicial power); Richard S. Higgins& Paul H. Rubin, Judicial Discretion, 9 J. LEGAL STUD. 129, 130 (1980).

106The rate of departure from the Guidelines was less than 15% in the early 1990s.107Recall that the Commission found that demographic differences under the mandatory Guide-

lines regime were lower during the PROTECT Act. See supra note 28.108The probability of reversal on sentencing matters fell from 36% in 2006 (under de novo review),

to 26% in 2008 (under abuse of discretion review). I calculate rate of appellate reversals using yearlydata on the universe of criminal appeals from the United States Sentencing Commission. Reversalis defined as all reversals and remands on appeals arising out of Booker sentencing issues.


under the previous Guidelines regime.109 Acculturation would suggest that judgeswith greater exposure to Guidelines sentencing would be less likely to depart fromthe Guidelines in the aftermath of Booker.

Another potential mechanism is anchoring, a type of cognitive bias in whichdecision-makers rely heavily on one piece of information and fail to make ratio-nal adjustments.110 Judge Nancy Gertner of the District of Massachusetts predictedthat the Guidelines would still play a predominant role for all judges post Bookerbecause “appellate courts have insisted that district court judges begin with - ef-fectively, ‘anchor’ their decisions - in the Guidelines before considering anythingelse.”111 Thus, to the extent that federal district judges are effectively anchored tothe Guidelines, one may not observe much deviation in sentencing practices afterBooker. Indeed, because district courts continued to calculate the applicable Guide-lines range in the aftermath of Booker, scholars commented in the year followingBooker that the federal sentencing system remained virtually unchanged.112

B. Sentencing Data

This Article provides the first comprehensive empirical evidence on the impact ofBooker and its progeny on inter-judge sentencing disparities. As noted previously,the Scott study is the only empirical study thus far, but is limited to the 2,262 casessentenced by judges who served continuously from 2001 to 2008 in the Bostoncourthouse of the District of Massachusetts.113 While the study is a first step incharacterizing the extent to which inter-judge sentencing practices have changed inthe aftermath of Booker, the Boston courthouse is likely unrepresentative of sen-

109See Judge Nancy Gertner, Supporting Advisory Guidelines, 3 HARV. L. POL’Y REV. 261,262 (2009) (“[A]fter twenty years of strict enforcement, the Federal Sentencing Guidelines havea gravitational pull on sentencing and are likely to shape the way judges view sentencing, evenif they are now only advisory. Indeed, the greatest danger is not that judges will exercise theirnew discretion, but that they will not.”); Stith, The Arc of the Pendulum, supra note 37, at 1496-97(“incumbent sentencing decision makers may be reluctant to regard as unreasonable the sentencesthey were obliged to seek and impose for two decades”); Frank O. Bowman, Beyond Band-Aids: AProposal for Reconfiguring Federal Sentencing After Booker, 2005 CHI. LEGAL F. 149, 187 (2005)(arguing that advisory guidelines might still constrain judicial discretion “if for no other reason thanthat the federal bench has become acculturated to the Guidelines over the last seventeen years”).

110See, e.g., Amos Tversky & Daniel Kahneman, Judgment Under Uncertainty: Heuristics andBiases, 185 SCIENCE 1124, 1130-31 (1974); see also Birte Englich et al., Playing Dice with Crimi-nal Sentences: The Influence of Irrelevant Anchors on Experts’ Judicial Decision Making, 32 PER-SONALITY & SOC. PSYCHOL. BULL. 188, 188 (2006) (experimental results showing that criminalsentences were higher if participants were confronted with a randomly high rather than a low an-chor).

111See Judge Nancy Gertner, What Yogi Berra Teaches About Post-Booker Sentencing, 115 YALEL.J. POCKET PART (2006) 137, 138-40 (describing the Guideline Manual as a “ready-made an-chor”).

112See Douglas A. Berman, Tweaking Booker: Advisory Guidelines in the Federal System, 43HOUSTON L. REV 341, 349 (2006) (“a culture of guideline compliance has persisted after Booker”).Berman also suggests that Commission data in the year after Booker indicates that “federal sentenc-ing judges are exercising their new discretion relatively sparingly.” Id. at 351.

113Scott, supra note 27, at 17.


tencing patterns in other courthouses across the United States.114 As a result, thepresence of growing inter-judge sentencing disparities after Booker in the Bostoncourthouse may be the result of the particular caseload and judicial compositionof that court, making conclusions that Booker has increased inter-judge sentencingdisparities likely not generalizable across other courts. In fact, a comparison of theBoston courthouse to other district courthouses reveals that the Boston courthouseexperienced a greater increase in inter-judge disparities following Booker than theaverage court in the nation.

Prior empirical research on inter-judge disparity and the impact of judicial de-mographics on sentencing practices has been hampered by the lack of judge identi-fiers in available data.115 Because cases are generally randomly assigned to judgeswithin a district courthouse, judge identifiers allow one to compare judges withinthe same court and in the same time period, capturing judge differences in sentenc-ing rather than different caseloads.116 However, the Sentencing Commission datadoes not identify the sentencing judge.117 In response, most researchers have re-sorted to using aggregate district-level variation in judicial demographics to controlfor judge sentencing preferences,118 but this methodology is flawed if districts withdifferent judicial compositions differ in ways in that affect all judges within the dis-trict.119 A few researchers have resorted to hand matching sentencing data from theSentencing Commission with Public Access to Court Electronic Records (PACER),but due to the intensive matching process, sample sizes have been severely lim-ited.120

This paper is the first in over 25 years to match sentencing data from fiscal years2000-2009 to judge identifiers in all 94 district courts, allowing for a comprehensivelook at inter-judge sentencing disparities after Booker. In order to overcome the lackof judge identifiers in sentencing data, I utilize data from three sources: The UnitedStates Sentencing Commission, the Transactional Records Access Clearinghouse,and the Federal Judicial Center. I describe each dataset in turn.

United States Sentencing Commission - I use data from the United States Sen-tencing Commission (USSC) on records of all federal offenders sentenced pursuant

114See supra note 36 for a discussion of the large geographical differences in sentencing andGuidelines treatment.

115The Anderson et al. study is the only empirical work with comprehensive sentencing data withjudge identifiers in the past 25 years. See Anderson et. al, supra note 10, at 287.

116According to the Administrative Office of the United States Courts, “The majority of courtsuse some variation of a random drawing.”

117United States Sentencing Commission, GUIDE TO PUBLICATIONS & RESOURCES 2007-200845 (2007), available at http://www.ussc.gov/publicat/Cat2005.pdf (“Pursuant to the policy on publicaccess to Sentencing Commission documents and data, all case and defendant identifiers have beenremoved from the data.” (internal citation omitted)).

118Fischman & Schanzenbach, Strategic Judging, supra note 33 (relying only on district-levelvariation in observable characteristics of judges).

119For instance, a district with a greater percentage of Democratic judges could be different fromother districts. It may be that both Democratic and Republican judges within the district sentencedifferently from judges in other districts, so any effect cannot be solely attributable to Democraticjudges.

120See, e.g., Schanzenbach & Tiller, Reviewing the Sentencing Guidelines: Judicial Politics, Em-pirical Evidence, and Reform, 75 U. CHI. L. REV. 715 (2008); Scott, supra note 27, at 15-16(describing the PACER method used to match records to sentencing data).


to the Sentencing Guidelines and Policy Statements of the SRA of 1984 in fiscalyears 1994-2009 (October 1, 1993 - September 30, 2009). The USSC providesdetailed sentencing data on federal defendants, but defendant and judge identifiersare redacted.121 The dataset contains information from numerous documents on ev-ery offender: Indictment, Presentence Report, Report on the Sentencing Hearing,Written Plea Agreement (if applicable), and Judgment of Conviction.

Court characteristics include the district court and circuit in which sentencingoccurred, in addition to the sentencing month and year.122 Demographic variablesinclude defendant’s race, gender, age, citizenship status, educational attainment,and number of dependents. The primary offense variable is the primary offensetype for the case generated from the count of conviction with the highest statutorymaximum.123 Data is also available on whether the offense carries a mandatoryminimum sentence under various statutes, and whether departures from the statu-tory minimum are granted, either under a substantial assistance motion or applica-tion of the safety valve. Offense level variables include the base offense level, thebase offense level after Chapter Two adjustments and the final offense level afterChapter Three adjustments. Criminal history variables include whether the defen-dant has a prior criminal record (first time offender or prior offender), and whetherarmed career criminal status, career offender status, or repeat and dangerous sexoffender status is applied.124

For each offender, there is a computed Guidelines range, and a Guidelines rangeadjusted for applicable mandatory minimums. Sentencing outcomes include im-prisonment or probation, sentence length, and length of supervised release. Fromthese variables, I construct indicator variables for above range and below rangedepartures from the Guidelines.125

Transactional Records Access Clearinghouse - The Transactional Records Ac-cess Clearinghouse (TRAC) provides less comprehensive sentencing records ob-tained from detailed federal records and information from the Justice Departmentand the Office of Personnel Management. Defendant, offense characteristics andGuidelines application information is not included, but defendants are linked tosentencing judge.126

Federal Judicial Center - Finally, I obtain demographic information on fed-eral district court judges. Federal district judges are Article III judges who serve

121Over 90% of felony defendants in the federal criminal justice system are sentenced pursuant tothe SRA of 1984 and all cases are assessed to be constitutional.

122USSC data prior to 2004 includes information on the exact sentencing day, but this variable isnot available in later years. Information is also collected on the Guidelines amendment year used incalculations. All results are robust to controlling for amendment year, although the results presentedin this paper do not include this control.

123There are a total of 35 offense categories in the dataset. The most common offense is drugtrafficking, followed by immigration, fraud, firearms, and larceny.

124Data is also collected on the total number of criminal history points applied and the final crim-inal history category.

125An above range departure is defined as 1 for a defendant who received a sentence above themaximum Guidelines recommended range. Similarly, a below range departure is defined as 1 for adefendant who received a sentence below the minimum Guidelines recommended range.

126TRAC has compiled records on the criminal cases and the civil matters handled by federal dis-trict court judges in each of the 94 federal judicial districts through over 20 years of FOIA requests.


life term tenures. New appointments are generally made when a judge retires ordies.127 As of the current day, there are a total of 677 authorized Article III dis-trict judgeships.128 The number of federal district judgeships in each district courtvaries. The largest district court is the Southern District of New York with 28 au-thorized judgeships. The majority of other district courts have between 2-7 districtcourt judgeships. I collect information on judge race, gender, affiliation of appoint-ing president, tenure, whether the judge was appointed prior to the adoption of theGuidelines in 1987, and whether the judge was appointed after Booker.

C. Matching

In order to connect defendants to judges, I match observations across these threedatasets. First, I match sentencing records from the USSC to TRAC data. By districtcourt, matching is conducted on several key variables that can uniquely identifyeach record: sentencing year, sentencing month, offense type,129 sentence lengthin months, probation length in months, amount of monetary fine, whether the caseended by trial or plea agreement, and whether the case resulted in a life sentence.130

Then, I match the USSC and TRAC combined data to judge biographical data fromthe Federal Judicial Center. I successfully match over 90% of all USSC cases fromfiscal years 2000-2009, resulting in a matched dataset of 636,063 cases representing91 district courts (hereinafter “full sample”).131

D. Testing for Random Assignment

In an ideal experiment to test the impact of Booker, one would randomly allocate thetreatment - sentencing under an advisory Guidelines - to certain groups of judges.In this hypothetical, a group of judges within each district court would be randomlyassigned to the treatment group, while the others would comprise the control group.Because of the random assignment of the “Booker” treatment, any differences incaseload composition or judge characteristics would be on average the same acrossboth treatment and control groups. As a result, a straightforward comparison of thesentencing practices between judges in the treatment group (those who sentenceunder an advisory Guidelines), and the judges in the control group (those who sen-tence under a mandatory Guidelines), would capture the causal effect of greaterjudicial discretion via Booker on inter-judge differences in sentencing.

Unfortunately, this hypothetical experiment does not exist because the SupremeCourt’s decision in Booker applied to all judges. However, one can utilize the quasi-

127On a few occasions, Congress has also increased the number of judgeships within a district inresponse to changing population or caseload.

12867 positions are currently vacant. Authorized judgeships only refer to full-time non-seniorstatus judges.

129Data from USSC are coded to correspond with the offense categories in the TRAC data.130These match variables are comparable to those used by previous scholars under the PACER

method. See Schanzenbach & Tiller, supra note 120, at 729 (matching USSC data with PACERrecords on date and length of the sentence, and when necessary, the amount of any fine, the offensetype, and the Hispanic ethnicity of the defendant).

131The Federal Judicial Center does not collect demographic information on judges in three dis-tricts: Guam, Virgin Islands, and Northern Mariana Islands.


experiment created by the timing of the Booker decision. Assuming that judgeswithin the same district courthouse are randomly assigned cases from the same un-derlying caseload, one can compare inter-judge disparities before Booker to inter-judge disparities after Booker. If there are no other contemporaneous changes thataffect inter-judge sentencing, an increase in inter-judge disparities is attributable tochanges in judicial behavior, rather than underlying differences in case composi-tion.132 Moreover, random assignment of cases can also lead to estimates of therelationships between judicial demographics such as gender and experience, andsentencing practices.

The crucial assumption underlying the validity of the quasi-experiment is therandom assignment of cases to judges.133 According to the Administrative Officeof the United States Courts, “[t]he majority of courts use some variation of a randomdrawing” as prescribed by local court orders. While each district court is authorizedto specify its own methods of case assignment,134 “[m]ost district and bankruptcycourts use random assignment, which helps to ensure a fair distribution of cases andalso prevents ‘judge shopping,’ or parties’ attempts to have their cases heard by thejudge who they believe will act most favorably.”135

However, random assignment may be violated in some instances. For exam-ple, senior status judges with reduced caseloads may select the type of cases theyhear during the year,136 and some courts assign certain types of cases to particularjudges.137 Moreover, even if a court has local rules and orders that specify the use

132Previous research indicates that Booker, and in particular Rita, Gall, and Kimbrough did havea causal impact on judicial behavior, after controlling for unobservable year and monthly changesthat might affect sentencing. See Yang, supra note 34, at 17-18.

133As mentioned previously, the Anderson et al., Hofer et al., Scott, and TRAC studies all rely onrandom case assignment. See supra notes 10-14.

134Under 28 U.S.C. §137, “[t]he business of a court having more than one judge shall be dividedamong the judges as provided by the rules and orders of the court. The chief judge of the districtcourt shall be responsible for the observance of such rules and orders, and shall divide the businessand assign the cases so far as such rules and orders do not otherwise prescribe.” For example, in theArizona district court, “the Clerk must assign criminal cases to District Judges within each divisionby automated random selection and in a manner so that neither the Clerk nor any parties or theirattorneys will be able to make a deliberate choice of a particular Judge.” AZ. L. R. Crim. 5.1(a).In the Northern District of California, “[c]ases shall be assigned blindly and at random by the clerkby means of a manual, automated or combination system approved by the judges of the court.” CAGeneral Order No. 44. In Colorado,“[t]he clerk shall maintain a computerized program to assurerandom and public assignment of new cases on an equal basis among the judicial officers.” D.C.Colo. L Civ R 40.1.

135For a description of judge assignment methods, see the Federal Judicial Center FAQS, availableat http://www.fjc.gov/federal/courts.nsf/.

13628 U.S.C. §294 governs the assignment of cases to senior status judges. See, e.g., MA GeneralOrder 10-04 §4.2 (“A senior judge may limit the category of cases assigned to him or her or mayselect a special category of cases for assignment. For example, a senior judge may elect not to beassigned criminal cases or may elect to be assigned only patent cases.”)

137For instance, the Southern District of New York assigns civil and criminal cases such thatall judges, “except the chief judge, shall be assigned substantially an equal share of the categoriesof cases of the court over a period of time. There shall be assigned or transferred to the chiefjudge such matters as the chief judge is willing and able to undertake, consistent with the chiefjudge’s administrative duties.” Thus, assignment is based on equal caseload, rather than pure randomassignment, and the chief judge is exempted from the rules. See Rules for the Division of Business


of random assignment, empirically testing for random assignment is important be-cause random assignment can be violated if individuals seek to game the system.138

For instance, some courts have decried situations in which high profile cases aregiven to judges viewed as favorably disposed to one side.139

In order to dispose of potential violations to random case assignment, I empir-ically test for random assignment. To begin, I employ several sample restrictions.First, I drop judges who were formally retired prior to the beginning of the datasetin 2000 to remove the possibility of non-random assignment to senior status judgeswho continued to hear cases during the sample period. Second, I drop judges anddistrict courthouses with annual caseloads of less than 25 cases in order to obtain asufficient number of cases per judge for statistical power.140 Finally, I drop districtcourthouses with only one active judge.

With this restricted sample, I test for randomly assignment using the matchedUSSC, TRAC, and Federal Judicial Center data from 2000-2009.141 If cases are ran-domly assigned to judges, then judges should see on average, cases with the samedistribution of predetermined defendant characteristics. I test random assignmentbased on five fixed defendant characteristics: gender, age, a black race indicator,number of dependents, and an indicator for less than a high school degree.142 In-tuitively, there should be no significant correlation between a particular judge anddefendant characteristics if cases are randomly assigned. I drop all courthouses thatviolate random assignment, resulting in a subsample of 163 courthouses from 73district courts representing about 50% of the cases from 2000-2009, for a total of

Among District Judges, available at http://www.nysd.uscourts.gov/courtrules.php.138See J. Robert Brown Jr. & Allison Herren Lee, Neutral Assignment of Judges at the Court of

Appeals, 78 TEXAS L. REV. 1037 (2000) (describing nonrandom assignment in federal courts ofappeals).

139See United States v. Pearson, 203 F.3d 1243, 1264 (10th Cir. 2000) (“If a judge re-ceives case assignments not through some neutral system, but rather because of prosecutors’opinion that he or she is more favorably disposed to the government’s arguments than an-other judge in the same district, then a judge’s caseload might be based in part on pros-ecutors’ evaluations of judicial performance.”). For recent allegations of “gaming” the ran-dom assignment system, see William Safire, Essay; Norma The Plumber, N.Y. TIMES, July31, 2000, available at http://www.nytimes.com/2000/07/31/opinion/essay-norma-the-plumber.html(allegations that the Chief Federal Judge of the U.S. District Court (U.S.D.C.) went “off thewheel” to assign a politically sensitive case in a non-random fashion); see also Dan Fitzpatrick,Bank of America Manages to Avoid Judge Rakoff, WALL ST. J., May 17, 2010, availableat http://online.wsj.com/article/SB10001424052748703699804575247132437874588.html (non-random assignment of Bank of America case in the Southern District of New York).

140Results are robust to choice of caseload minimums, but follow the convention in prior literature.The Scott study excludes judges who imposed less than 25 sentences in a two year period. See Scott,supra note 27, at 17. Similarly, Anderson et. al exclude judges who sentence less than 30 casesduring a two year period. Anderson et al., supra note 10, at 287.

141See Appendix A for details.142For each of the five defendant characteristics, I regress the characteristic on district courthouse

by sentencing year fixed effects, sentencing month fixed effects and judge fixed effects. I test thehypothesis of no judge effects (the null hypothesis) using an F-test for whether the judge fixedeffects are equal to zero. I employ seemingly unrelated regression (SUR). For a discussion of theSUR technique, see David H. Autor & Susan N. Houseman, Do Temporary-Help Jobs ImproveLabor Market Outcomes for Low-Skilled Workers? Evidence from “Work First”, 2 AEJ: AppliedEconomics 96, 106-107 (2010). See Appendix A, Table A1.


270,334 cases (hereinafter “random sample”).

E. Trends in Sentencing

I first present graphical evidence of trends in sentence lengths and rates of belowrange departures over the time period 2000-2009. Graphical analyses confirm thatBooker did significantly alter the sentencing practices of judges. Figure 1 presentsa graph of average sentence lengths in months over time, with the timing of Bookerdelineated by the first vertical line and Kimbrough/Gall by the second vertical dash,along with predicted trend lines before and after Booker. Figure 1 indicates thatoverall sentence lengths were relatively stable in the five years prior to Booker, butbegan to decrease afterwards, particularly for cases in which a mandatory minimumwas not charged.143

FIGURE 1. SENTENCE LENGTHS IN MONTHS

3540

4550

5560

65Se

nten

ce L

engt

h in

Mon

ths

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Sentencing Year

All Sentences

2025

3035

4045

Sent

ence

Len

gth

in M

onth

s

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Sentencing Year

Excluding Mandatory Minimums

Notes: Data is from the full sample 2000-2009 on all incarcerated defendants. Data points aremonthly averages.

143Cases defined as excluding mandatory minimums are those in which a statutory mandatoryminimum was not charged, which represent over two-thirds of all cases.


Figure 2 presents a graph of the the average rate of below range departures from2000-2009, that is the percentage of cases in which the defendant receives a sen-tence below the Guidelines recommended minimum sentence. Figure 2 reveals atrend of decreasing rates of below range departures prior to Booker, characterizedby very low relative rates of departures in the PROTECT Act era. The decreas-ing trend in below range departures was significantly changed following Booker,which induced a sudden jump in the rate of departure, as well as an increasing trendthroughout Booker, Rita, Gall, and Kimbrough, back to pre-PROTECT era levels.

FIGURE 2. BELOW RANGE DEPARTURE RATES

.2.2

5.3

.35

.4.4

5.5

Rat

e of

Bel

ow R

ange

Dep

artu

res

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Sentencing Year

All Sentences

.2.2

5.3

.35

.4.4

5.5

Rat

e of

Bel

ow R

ange

Dep

artu

res

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Sentencing Year



Below range departures can be the product of both prosecutorial and judicialaction. To disentangle these two factors, Figure 3 presents trends in the averagerate of below range departures that are not a result of a government sponsored sub-stantial assistance motion.144 Figure 3 indicates a similar trend with respect to rates

144As noted previously, a substantial assistance motion by the government permits a departurefrom the Guidelines for a defendant who provides substantial assistance in the prosecution or inves-tigation of another person.


of non-government sponsored below range departures, suggesting that judicial be-havior has changed following the shift to an advisory Guidelines regime. However,while overall trends in sentencing have changed in the aftermath of Booker, andits progeny Rita, Gall, and Kimbrough, aggregate trends mask whether inter-judgevariation has increased.

FIGURE 3. NON-GOVERNMENT SPONSORED BELOW RANGE DEPARTURES

.15

.2.2

5.3

.35

.4.4

5R

ate

of B

elow

Ran

ge D

epar

ture

s

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Sentencing Year

All Sentences.1

5.2

.25

.3.3

5.4

.45

Rat

e of

Bel

ow R

ange

Dep

artu

res

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Sentencing Year



F. Measuring Inter-Judge Disparity: Analysis of Variance

To identify changes in inter-judge disparity, I employ an analysis of variance method-ology. Variants of this methodology has been used in the federal sentencing litera-


ture,145 as well as in the economics literature on teacher value added.146 The anal-ysis of variance technique measures inter-judge dispersion in sentencing outcomesbased on the variance of a judge-specific random variable.147

The analysis of variance technique assumes that the impact of a judge on sen-tencing outcomes is randomly and normally distributed within each district court-house such that the judge effect has mean = 0 and variance = σ2.148 For instance,suppose that there are n judges in a district courthouse. If the judges were iden-tical in their sentencing preferences, and cases are randomly assigned to judges,there would be no impact of a particular judge on sentencing outcomes. Each judgewould sentence in the exact same way and variation in the judge effect, as measuredby σ2, would equal zero. To the extent that judges do differ in their sentencingpractices based on personal ideologies or goals, one would observe a distribution ofjudge effects, as measured by the variance or standard deviation in judge effects, σ.The greater the inter-judge variation in outcomes, the larger the σ.

Analysis of variance allows one to estimate the standard deviation of judge ef-fects on sentence length, σ, after controlling for case and defendant characteristics.A finding that σ = 5 implies a defendant who is assigned to a judge that is onestandard deviation “harsher” than the average judge receives a five month longersentence. In order to capture changes in inter-judge disparity, this paper measuresσ in periods before Booker and after Booker. An increase in σ after Booker impliesan increase in inter-judge sentencing disparity after the Guidelines were rendered

145Studies using a similar methodology include Anderson et al., supra note 10; Joel Waldfogel,Aggregate Inter-Judge Disparity in Sentencing: Evidence from Three Districts, 4 FED. SENTENC-ING REP. 151 (1991); Abigail Payne, Does Inter-judge Disparity Really Matter? An Analysis of theEffects of Sentencing Reforms in Three Federal District Courts, 17 INT’L J. LAW & ECON. 337(1997).

146See, e.g. Raj Chetty et al., How Does Your Kindergarten Classroom Affect Your Earnings?Evidence from Project Star, 126 Q. J. ECON. 1593 (2011).

147This paper does not employ the statistical technique used by Scott. See Scott, supra note27. Scott regresses sentencing outcomes on dummy indicators for each sentencing judge, such thatthe corresponding R-squared measures the percentage of variance in the dependent variable that isexplained by sentencing judge identity. Scott, supra note 27, at 58. The author interprets an increasein the R-squared in time periods following Booker as indicative of growing inter-judge disparities.For instance, with regards to sentence lengths for cases excluding mandatory minimums, the authorfinds an increase in R-squared from 0.014 pre-Booker to 0.080 post Kimbrough/Gall. Id. at Table2, at 34. However, the R-squared measure is problematic for two main reasons. First, the measureof R-squared does not have a straightforward interpretation in terms of actual inter-judge variation,in contrast to a measure of the variance in a judge-specific random variable. Second, the magnitudeof an R-squared cannot be taken literally without some discussion of its statistical significance,which is proxied by the linear regression model’s significance. Scott’s linear regression models areoften statistically insignificant, suggesting that judge fixed effects are a poor predictor of sentencingoutcomes, but he does not qualify the magnitudes of the R-squared measures. For instance, themodel is statistically insignificant in two out of three of the studied periods in Table 2, and fourout of five of the studied periods in Table 3, at 34-40. In contrast, measures of inter-judge variancein an analysis of variance can be rigorously tested for statistical significance. Scott acknowledgesthis issue, noting that “[t]he fact that the model for the Kimbrough/Gall period is not significantreinforces the need for caution in interpreting the results for cases not governed by a mandatoryminimum.... Although the relationship in the Kimbrough/Gall period is strongly positive, the modelfalls well short of statistical significance.” Scott, supra note 27, at 34-35 FN 177.

148See Appendix B for details on the empirical methodology.


advisory. In particular, I separate the sample timeframe of 2000-2009 into fourmain periods: (1) Koon (October 2000-April 2003), (2) PROTECT Act (May 2003- January 2005),149 (3) Booker (January 2005 - November 2007), and (4) Kim-brough/Gall (December 2007 - September 2009).150

III RESULTS ON INTER-JUDGE AND REGIONAL DISPARITIES

A. Sentence Length

The following graphs present boxplots of the average sentence length imposed byeach judge relative to the average sentence length of the district courthouse in whichthe judge sits.151 A boxplot captures the distribution of judge sentencing practices,with a more narrow spacing of the boxplot evidencing less inter-judge disparity.In particular, the top and bottom of the box capture the spread between the 75percentile and 25 percentile mean judge effect, also known as the interquartile range(IQR). More extreme judge effects are represented in the whiskers of the boxplot,as well as by outliers. The greater the IQR and presence of outliers, the larger theinter-judge disparity within district courts.

The first panel of Figure 4 indicates that over 50% of judges are sentencingwithin a few months of the average courthouse mean, with some outliers in bothdirections. However, the spread of the distributions over the four time periods indi-cates an increase in the distribution of judge average sentence lengths relative to thecourt mean following Kimbrough/Gall. The spread between the 25th and 75th per-centile (IQR) expands modestly but noticeably across the time periods. FollowingKimbrough/Gall, the number of outliers also increases.

Note, however, that some of the inter-judge disparities may be attributable touneven applications of mandatory minimums by prosecutors. The second panel ofFigure 4 presents distributions of average judge sentences for those cases in which amandatory minimum was not charged, approximately two-thirds of all cases. Thesecases better represent disparities more likely attributable to judicial behavior. Asthe figure shows, following Kimbrough and Gall, judge sentence lengths begin todepart more radically from court averages, with substantially more outlier sentencelengths on both sides of the distribution.

149The PROTECT Act became effective as of April 30, 2003.150Booker was decided on January 12, 2005, and Kimbrough/Gall were decided on December 10,

2007. Although the USSC data only contains information on sentencing month and year, the datais coded to denote which January 2005 cases were pre and post Booker and which December 2007cases were pre and post Kimbrough/Gall.

151Case composition likely varies across district courts. Thus, I use a measure of judge sentencelength that is comparable across all districts, which can be accomplished by calculating averagesentence length by judge relative to the mean district courthouse sentence.


FIGURE 4. AVERAGE JUDGE SENTENCE LENGTHS IN MONTHS

-100

-50

050

100

150

Mea

n Ju

dge

Devia

ton

From

Cou

rt M

ean

Koon PROTECT Act Booker Kimbrough/Gall

All Sentences

-50

050

100

Mea

n Ju

dge

Devia

ton

From

Cou

rt M

ean



Notes: Data is from the random sample 2000-2009.

Statistical analysis of variance confirms the graphical patterns. Table 1 presentsa measure of σ for sentence lengths, the causal impact of being randomly assigneda one standard deviation “harsher” judge in the sentencing district courthouse.152

Each measure of σ is also accompanied by a 95% confidence interval, which in-dicates the statistical probability that the true measure of σ lies within the intervalrange. During the Koon period in which the Guidelines were binding and judgeswere governed by an abuse of discretion standard of appellate review, a defendantassigned to a “harsher” judge received a 2.6 month longer sentence than similardefendants sentenced by an average judge in the courthouse. By the time of thePROTECT Act, a defendant randomly assigned to a harsher judge received almosta 4 month longer sentence. Inter-disparities increased further following Booker.Booker and Kimbrough/Gall induced almost a doubling of inter-judge disparitycompared to the Koon period. A harsher judge sentenced a defendant 4.6 monthslonger than the court average in the immediate aftermath of Booker and over 5.2months longer after Kimbrough and Gall.

152Table 1 analysis includes all defendants that received a prison sentence, excluding those whoreceived probation.


The second panel of Table 1 excludes from the analysis those cases in which amandatory minimum was charged. During Koon, a one standard deviation “harsher”judge sentenced a defendant to 1.6 months more than the court average, and 3.2months longer during the PROTECT Act. Interestingly, inter-judge disparity fornon-mandatory minimum cases falls to 2.4 months during Booker, rising backup to 3.3 months after Kimbrough and Gall. Changes in σ are not significantfrom the PROTECT Act to Kimbrough and Gall, suggesting that on average, inter-judge disparities in sentence lengths of non-mandatory minimum cases have notchanged starkly during this period. However, inter-judge disparities are signifi-cantly larger following Gall/Kimbrough compared to the Koon period, more thandoubling. This evidence suggests that even in cases in which mandatory minimumswere not charged, inter-judge disparities have increased significantly.

Interestingly, the estimates of σ in the bottom panel are almost halved com-pared to those presented in the top panel where all sentences are included. Thisfinding suggests that a large proportion of inter-judge disparities may be driven bythe disparate application of mandatory minimums by prosecutors.

TABLE 1. INTER-JUDGE VARIATION IN SENTENCE LENGTHS

ALL SENTENCES

Period σ Lower bound Upper bound No. Obs.Koon 2.616 1.951 3.509 45407PROTECT Act 3.975 3.118 5.068 38316Booker 4.651 3.923 5.513 68129Kimbrough/Gall 5.282 4.519 6.175 48605

EXCLUDING MANDATORY MINIMUMS

Period σ Lower bound Upper bound No. Obs.Koon 1.599 1.115 2.293 30193PROTECT Act 3.217 2.633 3.931 27571Booker 2.392 1.903 3.008 46160Kimbrough/Gall 3.296 2.783 3.902 33732Notes: Data is from the random sample from 2000-2009. All regressions contain demographic con-trols, and controls for offense type, offense level and criminal history. Regressions also contain sen-tencing year fixed effects, sentencing month fixed effects, and district courthouse fixed effects.

Table 2 presents evidence of inter-judge variation in the decision to incarcer-ate, disentangling the decision to incarcerate from the sentence length decision.153

Given that Booker rendered the Guidelines advisory, a judge could potentially im-pose no prison sentence on a defendant, even if the Guidelines recommended mini-mum was non-zero. Indeed, Table 2 reveals that inter-judge disparity has increasedthroughout the four time periods, and most significantly following Kimbrough andGall. During Koon, a one standard deviation “harsher” judge was 2.9% more likely

153I define incarceration as a binary indicator, where 1 indicates that the defendant has received asentence, and 0 indicates no sentence imposed. Defendants who do not received a prison sentenceoften pay fines and serve probationary periods.


to incarcerate than the courthouse average. The effect increased to 3.3% during thePROTECT Act, 3.5% during Booker and almost 5% following Kimbrough/Gall.

TABLE 2. INTER-JUDGE VARIATION IN INCARCERATION RATE

ALL SENTENCES

Period σ Lower bound Upper bound No. Obs.Koon .0291 .0239 .0353 51122PROTECT Act .0335 .0271 .0414 41713Booker .0349 .0297 .0412 73782Kimbrough/Gall .0499 .0433 .0575 52586Notes: Data is from the random sample from 2000-2009. All regressions contain demographic con-trols, and controls for offense type, offense level and criminal history. Regressions also contain sen-tencing year fixed effects, sentencing month fixed effects, and district courthouse fixed effects.

B. Below Range Departures

Figure 5 shows the boxplots of average rates of below range departures by judge,relative to the district courthouse mean, for all incarcerated defendants. While therate of below range departures in the aggregate was lowest during the PROTECTAct period (Figure 2), the distribution of judge below range departure rates doesnot appear to be significantly different between Koon and the PROTECT Act. Infact, there are far fewer outliers during the first two years following Booker. How-ever, inter-judge deviations from the court mean expand visibly following Kim-brough/Gall. Figure 5 suggests that increasing inter-judge disparities in sentencelength as described in Part IV.A are partly attributable to growing inter-judge dis-parities in the rate of below range departures.


FIGURE 5. AVERAGE JUDGE RATES OF BELOW RANGE DEPARTURES

-.50

.51

Mea

n Ju

dge

Devia

ton

From

Cou

rt M

ean


All Below Range Departures

-1-.5

0.5

1M

ean

Judg

e De

viato

n Fr

om C

ourt

Mea

n




Table 3 confirms these graphical trends. The top panel of Table 3 presents re-sults including all sentences. During the Koon period, a defendant who was as-signed to a judge one standard deviation more “lenient” than the average judge was4.8% more likely to be sentenced below the Guidelines recommended minimum.154

During the PROTECT Act, a similar judge was 4.2% more likely to sentence be-low range. Following Booker, the “lenient” judge’s practices deviated more greatlyfrom the courthouse average, with a 5.4% rate immediately following Booker and6.7% rate after Kimbrough/Gall. The increased likelihood of below range depar-tures following Kimbrough/Gall is statistically significant from the Koon-era rate,revealing markedly higher inter-judge disparities.

Excluding cases with mandatory minimums reveals a very similar trend, withthe one standard deviation more “lenient” judge being 5.1% more likely to sentencebelow range during Koon, rising to 7.1% following Kimbrough/Gall. Note that themagnitudes of σ when all sentences are included (top panel), and when mandatoryminimums are excluded (bottom panel), are very similar. This finding suggests

154Here, I define a judge who sentences defendants at greater rates below range as more “lenient.”Leniency is used solely to connote a tendency to impose shorter sentence lengths.


that inter-judge disparities in below range departures are real and substantial, andnot the mere product of mandatory minimums. If anything, measures of inter-judgedisparity are lower in most periods when mandatory minimums are included. Recallthat a mandatory minimum that exceeds the Guidelines recommended minimumtrumps the Guidelines minimum, becoming the statutorily binding minimum, thuspotentially reducing inter-judge disparity. The findings suggest that the applicationof mandatory minimums may yield the appearance of inter-judge consistency.155

TABLE 3. INTER-JUDGE VARIATION IN BELOW RANGE DEPARTURES

ALL SENTENCES

Period σ Lower bound Upper bound No. Obs.Koon .0477 .0396 .0576 44338PROTECT Act .0427 .0337 .0542 36613Booker .0538 .0459 .0632 64781Kimbrough/Gall .0668 .0576 .0774 45267



Notes: Data is from the random sample from 2000-2009. All regressions contain demographic controls, andcontrols for offense type, offense level and criminal history. Regressions also contain sentencing year fixedeffects, sentencing month fixed effects, and district courthouse fixed effects.

Not only do mandatory minimums confound the accurate determination of inter-judge disparities, so do below range departures that are government sponsored. Ta-ble 4 analyzes inter-judge variation in only those below range departures that arejudicially initiated, rather than the result of a government substantial assistancemotion. Table 4 indicates that the increasing inter-judge disparities in below rangedepartures evidenced in Table 3 persist in this subset of departures. Inter-judge dis-parities increased from 4.3% during Koon to 5.9% after Booker and over 7.4% afterKimbrough/Gall, with the lowest inter-judge disparities during the PROTECT Act.Inter-judge disparities similarly increased throughout the period for the subset ofcases not subject to a mandatory minimum, from 4.3% during Koon to over 7.4%after Kimbrough/Gall. These results indicate that in the subset of departures that aremost likely attributable to judicial behavior, the PROTECT Act was not only asso-ciated with the lowest aggregate rates of downward departures, but also the lowestinter-judge disparities.

155See Scott, supra note 27, at 26 (“mandatory minimums may interfere with accurate assessmentof inter-judge sentencing disparity by creating the illusion of inter-judge consistency.”).


TABLE 4. INTER-JUDGE VARIATION IN BELOW RANGE DEPARTURES

NON-GOVERNMENT SPONSORED DEPARTURES

ALL SENTENCES




Notes: Data is from the random sample from 2000-2009. All regressions contain demographic con-trols, and controls for offense type, offense level and criminal history. Regressions also contain sen-tencing year fixed effects, sentencing month fixed effects, and district courthouse fixed effects.

C. Above Range Departures

Inter-judge disparities have increased not only in the rate of below range departures,but also the rate of above range departures, which comprise approximately 5% ofcases. Figure 6 presents the distribution of average rates of above range departuresby judge, relative to their district courthouse mean, for all incarcerated defendants.The graphs reveal an expansion in the distribution of above range departure rateswithin district courts, particularly between the 25th and 75th percentile of the dis-tribution. Increased inter-judge deviations are also reflected in the rate of aboverange departures for cases with no mandatory minimums charged. Although thereappear to be fewer extreme outliers following Kimbrough/Gall, the spread betweenthe 25th and 75th percentile is visibly larger compared to pre-Booker spreads.


FIGURE 6. AVERAGE JUDGE RATES OF ABOVE RANGE DEPARTURES

-.2-.1

0.1

.2.3

Mea

n Ju

dge

Devia

ton

From

Cou

rt M

ean


All Above Range Departures

-.2-.1

0.1

.2.3

Mea

n Ju

dge

Devia

ton

From

Cou

rt M

ean




Table 5 presents measures of inter-judge variation from the analysis of vari-ance and reveals significant and substantial increases in inter-judge disparities inabove range departures. In the top panel where all sentences are analyzed, a onestandard deviation “harsher” judge was 1.6% more likely to sentence a defendantabove range compared to the average judge in the courthouse during Koon. Whileinter-judge variation did not change substantially from Koon to the PROTECT Actperiod, to the first few years after Booker, this harsher judge was over 2.7% morelikely to sentence a defendant above range after Kimbrough/Gall. Inter-judge dis-parities in above range departures increased by over 70% from the beginning to theend of the time period.

Of course, mandatory minimums may explain a sizable fraction of defendantsthat are sentenced above range if the mandatory minimum trumps the maximumGuidelines recommended sentence. When cases with mandatory minimums are ex-cluded, inter-judge disparities are approximately halved. During Koon, inter-judgedisparities in above range departures were minimal, with a one standard deviation“harsher” judge being only .07% more likely to sentence above range. However,inter-judge disparities doubled by the end of the time period, to 1.3% after Kim-


brough/Gall, suggesting that increases in above range inter-judge disparities arenot the mere byproduct of mandatory minimums.

TABLE 5. INTER-JUDGE VARIATION IN ABOVE RANGE DEPARTURES

ALL SENTENCES




Notes: Data is from the random sample from 2000-2009. All regressions contain demographic controls, andcontrols for offense type, offense level and criminal history. Regressions also contain sentencing year fixedeffects, sentencing month fixed effects, and district courthouse fixed effects.

D. Sentencing Practices by Judge Demographics

The previous section finds that inter-judge disparities in sentence length, belowrange departures, and above range departures have increased significantly followingBooker, in particular after Kimbrough/Gall. In this section, I analyze whether in-creases in inter-judge disparities are idiosyncratic, resulting from all judges chang-ing their behavior in similar ways, or if judges are systematically differing fromtheir colleagues based on observable traits.156 Recall that due to the random as-signment of cases to judges within a district courthouse, any difference in judgesentencing practices can be solely attributable to a judge effect.

Consistent with previous research, I find significant and systematic differencesin the sentencing practices of both Democratic judicial appointees compared to theirRepublican appointed peers, and female judges compared to male judges.157 Thesedifferences magnified in the aftermath of Booker and Kimbrough/Gall, suggestingthat they are some of the sources of the growing inter-judge disparities identifiedearlier. The coefficients presented in Table 6 represent the sentencing tendency of aparticular type of judge compared to his or her colleagues within the same districtcourthouse, for an observably identical defendant and case, sentenced in the samemonth-year.

156See Appendix C.157See supra notes 30-32.


TABLE 6. SENTENCING PRACTICES BY JUDGE CHARACTERISTICSALL SENTENCES

(1) (2) (3) (4)Sentence Below Range Below Range Above Range

Non-GovtPost Booker 1.010 0.0211 0.0462** 0.0242***

(1.439) (0.0240) (0.0207) (0.0091)Tenure -0.0556 0.0005 0.0003 -0.0005*

(0.0446) (0.0009) (0.0009) (0.0003)Tenure*Booker 0.0461 -0.0006 0.0002 0.0010**

(0.0614) (0.0011) (0.0009) (0.0004)Tenure*Kimbrough 0.0126 0.0011 0.0023 0.0009*

(0.0672) (0.0019) (0.0019) (0.0005)Democratic -0.522 0.0109 0.0066 0.0016

(0.438) (0.0093) (0.0082) (0.0027)Democratic*Booker -0.609 0.0084 0.0086 0.0019

(0.571) (0.0082) (0.0081) (0.0037)Democratic*Kimbrough -0.756 0.0223* 0.0269** -0.0027

(0.731) (0.0115) (0.0120) (0.0051)Female 0.537 -0.0025 -0.0073 0.0051

(0.468) (0.0126) (0.0132) (0.0042)Female*Booker -1.692*** 0.0155 0.0164 -0.0100*

(0.503) (0.0121) (0.0138) (0.0054)Female*Kimbrough -0.401 0.0110 0.0145 0.0012

(0.576) (0.0192) (0.0203) (0.0055)Black -0.873 0.0192 0.0266 -0.0082*

(0.581) (0.0161) (0.0173) (0.0044)Black*Booker -0.624 -0.0106 -0.0104 0.0066

(0.951) (0.0176) (0.0206) (0.0071)Black*Kimbrough 0.459 -0.0334 -0.0367 0.0137

(1.405) (0.0262) (0.0315) (0.0095)Pre Guide 0.561 -0.0266** -0.0218 0.0044

(0.831) (0.0134) (0.0135) (0.0053)Pre Guide*Booker -0.765 0.0254 0.0091 -0.0137**

(0.952) (0.0179) (0.0176) (0.0069)Pre Guide*Kimbrough -0.418 0.0129 -0.0027 -0.0003

(1.224) (0.0308) (0.0313) (0.0099)Booker 0.276 0.0039 -0.0063 -0.0057

(0.735) (0.0139) (0.0132) (0.0050)Kimbrough -0.443 0.0038 -0.0076 -0.0017

(1.193) (0.0191) (0.0182) (0.0124)Observations 206,292 205,160 178,150 205,160R-squared 0.785 0.217 0.281 0.083Notes: Data is from the random sample from 2000-2009. All regressions contain controls for offense type, anddummies for each offense level and criminal history combination. Regressions also contain district courthouse fixedeffects, sentencing year and sentencing month fixed effects, and standard errors are clustered at the district courthouselevel. *** = significant at 1 percent level, ** = significant at 5 percent level, * = significant at 10 percent level.


Column 1 presents results for sentence length. Female judges significantly al-tered their practices from their male counterparts within the same courthouse. Im-mediately after Booker, female judges sentenced observably similar defendants toapproximately 1.7 months less than their male colleagues.158 Columns 2 presentsresults for the rate of below range departures. Inter-judge disparities in rates ofbelow range departures appear to be somewhat attributable to differences by judgepolitical affiliation. Following Kimbrough/Gall, Democratic judicial appointees aresignificantly more likely to depart downwards from the Guidelines recommendedrange, compared to their Republican appointed colleagues. For a similar defendantand crime, Democratic judges were 2.2% more likely to depart downwards. In-terestingly, pre-Guidelines (1987) appointees are significantly less likely to departdownwards from the Guidelines throughout the entire 2000-2009 period. Inter-judge differences by demographics are also prominent for the subset of below rangedepartures not sponsored by the government, as seen in column 3. Democratic ju-dicial appointees are 2.7% more likely to depart downwards compared to their Re-publican colleagues. Judges appointed post Booker are almost 5% more likely todepart downwards compared to pre-Booker appointees.

Finally, column 4 presents results for above range departures. Following Booker,female judges are 1% less likely to sentence above range compared to their malecolleagues. Inter-judge differences also appear by judge tenure, defined as num-ber of years of experience sentencing under the mandatory Guidelines regime forthose judges appointed under the mandatory regime. In general, a judge withgreater years of experience under the mandatory regime is significantly less likelyto sentence above range, but this pattern reverses in the aftermath of Booker andKimbrough/Gall, where judges with greater experience are more likely to sentenceabove range. Black judges are in general .08% less likely to sentence above rangecompared to white judges, and pre-Guidelines appointees are 1.4% less likely todepart upwards after Booker. Also striking are the inter-judge differences gener-ated between judges appointed pre-Booker and judges appointed post Booker. Ingeneral, post Booker judicial appointees are 2.4% more likely to sentence aboverange their their pre-Booker appointed peers.

Table 7 presents the results excluding cases with mandatory minimums. Judgedifferences by political affiliation of appointing president and gender persist. Fol-lowing Kimbrough/Gall, Democratic appointees issue sentences 0.8 months shorterthan their Republican colleagues for observably similar defendants, are 3.1% morelikely to depart downwards, and 3.1% more likely to depart downwards in caseswithout government substantial assistance motions. Similarly, female judges issue0.8 month shorter sentences than their male colleagues following Booker. Differ-ences between pre-Booker and post Booker appointees also remain. Post Bookerare approximately 5% more likely to depart downwards in all cases, and over 7%more likely when there is no substantial assistance motion.

158The Booker indicator here represents only the period 2005-2007 prior to Kimbrough.


TABLE 7. SENTENCING PRACTICES BY JUDGE CHARACTERISTICSEXCLUDING MANDATORY MINIMUMS

(1) (2) (3) (4)Sentence Below Range Below Range Above Range

Non-GovtPost Booker -1.010 0.0496** 0.0713*** 0.0033

(0.683) (0.0242) (0.0228) (0.0061)Tenure -0.0278 2.90e-05 -4.93e-05 -0.0001

(0.0272) (0.0008) (0.0008) (0.0002)Tenure*Booker 0.0260 -0.0003 -0.0002 0.0002

(0.0343) (0.0010) (0.0009) (0.0003)Tenure*Kimbrough -0.0130 0.0022 0.0029 0.0006

(0.0454) (0.0019) (0.0021) (0.0005)Democratic -0.382 0.0077 0.0040 0.0011

(0.267) (0.0088) (0.0086) (0.0022)Democratic*Booker -0.291 0.0047 0.0069 -0.0023

(0.290) (0.0101) (0.0104) (0.0028)Democratic*Kimbrough -0.830* 0.0310** 0.0307** -0.0030

(0.456) (0.0141) (0.0152) (0.0044)Female 0.242 -0.0019 -0.0077 0.0019

(0.324) (0.0145) (0.0156) (0.0023)Female*Booker -0.811** 0.0145 0.0168 -0.0033

(0.407) (0.0158) (0.0168) (0.0032)Female*Kimbrough -0.217 0.0049 0.0123 0.0015

(0.501) (0.0219) (0.0221) (0.0047)Black -0.501 0.0297* 0.0309* -0.0031

(0.425) (0.0175) (0.0180) (0.0021)Black*Booker 0.220 -0.0258 -0.0227 0.0079*

(0.637) (0.0200) (0.0223) (0.0045)Black*Kimbrough 0.587 -0.0533* -0.0473 0.0009

(0.811) (0.0318) (0.0380) (0.0045)Pre Guide 0.256 -0.0233 -0.0229 0.0037

(0.400) (0.0152) (0.0157) (0.0033)Pre Guide*Booker -0.312 0.0166 0.0145 -0.0019

(0.566) (0.0195) (0.0200) (0.0049)Pre Guide*Kimbrough 0.364 -0.0073 -0.0137 -0.0017

(0.949) (0.0331) (0.0370) (0.0091)Booker 0.492 -0.0062 -0.0174 0.0041

(0.372) (0.0145) (0.0138) (0.0035)Kimbrough 0.184 -0.0201 -0.0277 -0.0030

(0.670) (0.0189) (0.0192) (0.0069)Observations 141,647 141,386 131,417 141,386R-squared 0.819 0.244 0.283 0.037Notes: Data is from the random sample from 2000-2009. All regressions contain controls for offense type, anddummies for each offense level and criminal history combination. Regressions also contain district courthouse fixedeffects, sentencing year and sentencing month fixed effects, and standard errors are clustered at the district courthouselevel. *** = significant at 1 percent level, ** = significant at 5 percent level, * = significant at 10 percent level.


Overall, these results suggest that sentencing differences associated with judgegender and political affiliation are magnified after Booker and/or Kimbrough/Gall.Such differences are likely sources of growing inter-judge disparities. Given theselarge changes in inter-judge disparities following Booker, judges do not appear tobe completely “anchored” to the Guidelines.159

However, the finding that post Booker judicial appointees are more likely to de-part downwards from the Guidelines than pre-Booker appointees is consistent witha story in which judges with no prior experience sentencing under the Guidelinesregime are less anchored.160 The “anchor” of the Guidelines sentence may be moreprominent to pre-Booker appointees because these judges are more acculturatedand experienced under the Guidelines. In contrast, the “anchor” is less prominentfor post Booker appointees. These potential anchoring differences between pre andpost Booker appointees suggests that defense lawyer James Felman’s predictionsmay be true - that disparities may “increase as the years go by and the bench isfilled with individuals who have no history with binding guidelines.”161 Yet, inter-judge disparities that are due to the entrance of new judges to the federal benchmight only reflect a short-term surge in disparity. As time goes on and all judgeshave no history with binding Guidelines, inter-judge disparities attributable to thissource may fall.

E. Regional Disparity: Inter-District Variation

Commentators have suggested that different political climates across districts andcircuits can affect sentencing practices,162 yielding empirical findings that jurisdic-tional effects are prominent in federal sentencing.163 The recent 2012 Commission

159Of course, a degree of anchoring is likely occurring, which indicates that these results are onlylower bound estimates on increases in inter-judge disparities in a system in which sentencing doesnot begin with the Guidelines calculation.

160In robustness checks, I find that the behavior of post Booker appointees in my data is not due tothe fact that they are George W. Bush appointees based on comparisons with pre-Booker George W.Bush appointees. Rather, sentencing behavior seems to be associated with lack of experience underthe binding Guidelines.

161See Felman, supra note 35, at 98-99.162See, e.g, Nora Demlietner, The Nonuniform Developments of Guideline Law in the Courts, 6

FED. SENT. REP. 239 (1994) (describing district and circuit specific “personas” in sentencing caselaw); Jeffrey T. Ulmer & John Kramer, Court Communities Under Sentencing Guidelines: Dilemmasof Formal Rationality and Sentencing Disparity. 34 CRIMINOLOGY 383, 402-403 (1996) (Based onan analysis of three county courts in Pennsylvania, the authors argue that local courts operate underformal sentencing standards articulated by a guidelines regime and substantive, extralegal factorsrelevant to local courts, such as “perceptions of the defendant’s characteristics, local concerns, andcourt actors’ organizational and individual interests.”); Jeffrey T. Ulmer, SOCIAL WORLDS OF SEN-TENCING: COURT COMMUNITIES UNDER SENTENCING GUIDELINES (1977).

163See Celesta Albonetti, Sentencing Under the Federal Sentencing Guidelines: Effects of De-fendant Characteristics, Guilty Pleas, and Departures on Sentence Outcomes for Drug Offenses,1991-1992, 31 LAW & SOC’Y REV. 789, 815-16 (1997) (finding significant circuit-specific sen-tencing practices for black and white defendants); Ronald Everett & Roger Wojtkiewicz, Difference,Disparity, and Race/Ethnic Bias in Federal Sentencing, 18 J. QUANTITATIVE CRIM. 189 (2002)(finding harsher sentencing in the southern circuits compared to other circuits); Paula Kautt, Lo-cation, Location, Location: Interdistrict and Inter circuit Variation in Sentencing Outcomes for


report finds that rates of non-government sponsored below range sentences increas-ingly depend upon the district court in which the defendant is sentenced and theinfluence of the Guidelines on sentence length varies significant by circuit court.164

However, some researchers have found that between-district variation in the effectsof extralegal factors on sentencing have not increased following Booker.165

Recall that the identification of the impact of Booker on inter-judge disparitywithin a district courthouse relies on the random assignment of cases to judges.Such random assignment does not exist between districts, such that differences indistrict sentencing practices are most likely also due to differing caseloads. For in-stance, the Commission has noted that simple comparisons of regional variationsmight be attributable to different types of crimes within the general offense cate-gories, such that frauds sentenced in the Southern District of New York are sub-stantially different from frauds sentenced in the District of North Dakota.166

While I cannot control for unobservable differences across districts, the em-pirical methodology in this Article does control comprehensively for observableoffender and case characteristics.167 For the inter-district results, I utilize the fullsample described in Section 3 as random assignment is no longer a prerequisite. Inthe context of inter-district disparities, analysis of variation now yields an estimateof the standard deviation of district effects on sentence length, σ, after controllingfor case and defendant characteristics. Thus, a finding of σ = 5 now suggests thata defendant sentenced in a one standard deviation “harsher” district is sentenced tofive more months in prison, than if he were sentenced in an average district court.

Figure 7 presents raw distributions of sentence lengths by circuit court, exclud-ing life sentences.168 While uncontrolled differences cannot be treated as regionaleffects because districts have very different case compositions, the raw data re-veals substantial differences in sentence length, both in the distribution between the25th percentile and 75th percentile, and presence of outliers. For instance, priorto Booker, defendants sentenced in the Third Circuit received an average sentenceof 56 months, compared to an average sentence of 78 months for defendants sen-tenced in the neighboring Fourth Circuit. After Booker, the average Third Circuitdefendant received 62 months in prison, while the average Fourth Circuit defendant

Federal Drug-Trafficking Offenses, 19 JUSTICE QUARTERLY 633, 659 (2002) (“despite the federalsystem’s congressionally mandated return to determinate sentencing, extra-legal factors (specificallyjurisdictional effects) continue to influence the federal sentencing system and its outcomes directlyand indirectly”).

164United States Sentencing Commission, REPORT ON THE CONTINUING IMPACT OF UNITEDSTATES V. BOOKER ON FEDERAL SENTENCING (2012).

165See Jeffery T. Ulmer et al., The “Liberation” of Federal Judges’ Discretion In the Wake ofthe Booker/Fanfan Decision: Is there Increased Disparity and Divergence Between Courts?, 28JUSTICE QUARTERLY 799 (2011).

166United States Sentencing Commission, FIFTEEN YEARS OF GUIDELINES SENTENCING: ANASSESSMENT OF HOW WELL THE FEDERAL CRIMINAL JUSTICE SYSTEM IS ACHIEVING THEGOALS OF SENTENCING REFORM at 99-100 (Nov. 2004) (“Similarly, variations in the rates of aparticular type of departure among different districts must be evaluated within a larger context ofeach district’s distinctive adaptation to the guidelines system. Inferring unwarranted disparity fromuncontrolled comparisons of average sentences or rates of departure may be erroneous.”).

167Nevertheless, the results on inter-district variation should be interpreted with some caution tothe extent that there are unobserved differences across district courts that cannot be captured.

168Life sentences are top coded as 470 months in the dataset.


received 84 months in prison.

FIGURE 7. DISTRIBUTION OF SENTENCE LENGTHS, BY CIRCUIT COURT

010

020

030

040

050

0Se

nten

ce L

engt

h in

Mon

ths

D.C.Firs

t

Secon

dThir

dFou

rth Fifth Sixth

Seven

thEigh

tNint

hTe

nth

Eleven

th

Sentences - Before Booker

010

020

030

040

050

0Se

nten

ce L

engt

h in

Mon

ths

D.C.Firs

t

Secon

dThir

dFou

rth Fifth Sixth

Seven

thEigh

tNint

hTe

nth

Eleven

th

Sentences - After Booker

Notes: Data is from the full sample 2000-2009.

Table 8 shows that after controlling for case and defendant characteristics, thereis substantial variation in the sentence that a defendant would receive dependingon which district court he is sentenced in. During the Koon period, a defendantsentenced in a one standard deviation “harsher” district court received a 7.8 monthlonger prison sentence. This inter-district disparity increased to 8.4 months duringthe PROTECT Act, to 10.4 months immediately following Booker, reaching an 11.3month difference after Kimbrough/Gall. By late 2007, inter-district disparities weresignificant larger than existed under Koon.

Analyzing the subset of cases in which a mandatory minimum was not chargedmore than halves the magnitude of σ, the measure of inter-district variation. Thelower panel of Table 8 indicates that a one standard deviation “harsher” district courtsentenced a defendant to 3.6 months longer than the average district court underKoon, 4.4 months longer after the PROTECT Act, 4.9 months longer after Bookerand 5.2 months longer after Kimbrough/Gall. Once again, inter-district variationis statistically greater after Kimbrough/Gall compared to Koon. Nevertheless, the


finding the the magnitude of inter-district variation is reduced by over half whena statutory minimum is not charged indicates that the application of mandatoryminimums is a large contributor to inter-district disparities, particularly in light ofthe fact that mandatory minimums represent only approximately one-third of thecases.

TABLE 8. INTER-DISTRICT VARIATION IN SENTENCE LENGTHS

ALL SENTENCES

Period σ Lower bound Upper bound No. Obs.Koon 7.799 6.701 9.077 159163PROTECT Act 8.439 7.232 9.849 83829Booker 10.397 8.961 12.063 148560Kimbrough/Gall 11.262 9.692 13.087 106033


Period σ Lower bound Upper bound No. Obs.Koon 3.578 3.057 4.188 104917PROTECT Act 4.403 3.758 5.159 58104Booker 4.920 4.229 5.724 97628Kimbrough/Gall 5.157 4.419 6.018 72036Notes: Data is from the full sample from 2000-2009. All regressions contain demographic controls,and controls for offense type, offense level and criminal history. Regressions also contain sentencingyear fixed effects, sentencing month fixed effects.

Table 9 reveals that district courts also significantly differ in their rates of belowrange departures. A defendant sentenced in a one standard deviation more “lenient”district is 12.1% more likely to be sentenced below the Guidelines range, comparedto the average district court, during Koon. This measure of inter-district variationfor below range departures remains relatively constant throughout the entire sample,both including and excluding mandatory minimums. Booker and Kimbrough/Galldo not appear to have dramatically increased inter-district disparity with regards todownward departures.


TABLE 9. INTER-DISTRICT VARIATION IN BELOW RANGE DEPARTURES

ALL SENTENCES




Notes: Data is from the full sample from 2000-2009. All regressions contain demographic controls, and controls for offensetype, offense level and criminal history. Regressions also contain sentencing year fixed effects, sentencing month fixedeffects.

F. Prosecutorial Contributions to Disparities

Prosecutors likely contribute to observed inter-judge disparities through their charg-ing decisions. One area of great prosecutorial discretion is the decision to charge anoffense that carries a mandatory minimum. Justice Breyer has stated that manda-tory minimum statutes “transfer sentencing power to prosecutors, who can deter-mine sentences through the charges they decide to bring.”169 Strategic charging ofmandatory minimums are likely more prominent after Booker as some prosecutorscharge mandatory minimums in order to narrow a judge’s discretion.170

In a 2011 Congressional report on mandatory minimum penalties, the Sentenc-ing Commission found significant variation in the extent which prosecutors ap-plied enhancements for mandatory minimum penalties under drug trafficking of-fenses.171 The report documented over 75% of eligible defendants receiving thestatutory mandatory minimum penalty in some districts, but none of eligible defen-dants in other districts receiving the enhancement.172 Furthermore, recent work byresearchers shows evidence of significant racial disparities in prosecutorial charg-ing.173

169Harris v. United States, 536 U.S. 545, 571 (2002) (Breyer, J., concurring).170See Testimony of Patrick J. Fitzgerald, U.S. Attorney, Northern District of Illinois, to the United

States Sentencing Commission, at 252 (Sept. 2009) (“[A] prosecutor is far less willing to foregocharging a mandatory minimum sentence when prior experience shows that the defendant will ulti-mately be sentenced to a mere fraction of what the guidelines range is.”).

171United States Sentencing Commission, REPORT TO CONGRESS: MANDATORY MINIMUMPENALTIES IN THE FEDERAL CRIMINAL JUSTICE SYSTEM, supra note 85, at 252-261.

172Id. at 111-113 (prosecutors reported wide variations in the district practices on seeking statu-tory minimum penalties).

173See M. Marit Rehavi & Sonja B. Starr, Racial Disparity in Federal Criminal Charging and ItsSentencing Consequences, University of Michigan Law & Economics Working Paper 2012. Usingdata on 58,000 federal criminal cases from 2007-2009, the authors find significant racial dispari-ties in severity of initial charges. In particular, they find that black offenders are on average more


Prosecutors are also in charge of the decision to reduce sentences below themandatory minimum if the defendant offers “substantial assistance” during anotherinvestigation or prosecution.174 If the government files a motion for substantialassistance for a case involving a mandatory minimum sentence, the court has thepower to impose a sentence as low as probation.175 Scholars have commented thatthe substantial assistance departure provision affords prosecutors immense discre-tion over both plea bargaining and sentencing outcomes under the Guidelines.176

I find that the application of mandatory minimums appears to be a large con-tributor to inter-judge disparities. Given the random assignment of cases to judgeswithin a district courthouse, equal application of mandatory minimums among el-igible cases prior to assignment would result in no significant judge differencesin the rate of mandatory minimums applied. However, mandatory minimums canalso be charged after assignment through the use of superseding indictments, givingprosecutors even greater control in their charging decisions. The results in Table 10reveal small, but significant differences in the percentage of cases with applicablemandatory minimums across judges. A judge one standard deviation out in the dis-tribution was 2.3% more likely to see a case with a mandatory minimum duringKoon, but 3.5% more likely after Kimbrough/Gall. The increase in the differen-tial rates of mandatory minimums after Kimbrough/Gall coincide with substantialincreases in inter-judge disparities in below range departures. These results are con-sistent with a story in which prosecutors are attempting to rein in judicially induceddownward departures through the use of mandatory minimums.

than two times as likely to be subjected to a mandatory minimum sentence compared to similarwhite offenders, and that a major part of the racial gap in sentence length can be attributed to theprosecutorial bias in initial charge.

17418 U.S.C. §3553(e). A judge has some leeway in reducing sentence length for certain drugtrafficking offenses under the “safety valve” provision, which allows a judge to reduce the punish-ment for low level, first time offenders. See 18 U.S.C. §3553(f). Report to Congress: MandatoryMinimum Penalties in the Federal Criminal Justice System, supra note 85, states that in recent years,white defendants in drug cases are more frequently granted the safety valve exception compared toother defendants.

175According to the Sentencing Commission, substantial assistance motions reduce the averagedefendant’s sentence length by 50%.

176See Nagel & Schulhofer, supra note 37, at 550 (“The use of the section 5K1.1 substantial-assistance motion varies from jurisdiction to jurisdiction....There is no limit on the amount of reduc-tion once the motion is submitted. The section 5K1.1 motion is also used to avoid guideline rangesor mandatory minimum sentences for sympathetic defendants – even when there has been no gen-uine substantial assistance.”); Michael H. Tonry, SENTENCING MATTERS (1996); Jeffrey StandenPlea Bargaining in the Shadow of the Guidelines, 81 CALIFORNIA L. REV. 1471 (1993).


TABLE 10. INTER-JUDGE VARIATION IN MANDATORY MINIMUMS

ALL SENTENCES



TABLE 11. INTER-JUDGE VARIATION IN SUBSTANTIAL ASSISTANCE

ALL SENTENCES



Appendix A Table A2 and Table A3 confirm that a large portion of inter-districtdifferences in the sentencing of observably similar defendants arises from districtvariation in both the charging of mandatory minimums and the application of asubstantial assistance motion. Table A2 reveals that a defendant sentenced in aone standard deviation “harsher” district is approximately 6% more likely to becharged with a mandatory minimum. Table A3 also presents evidence of largeinter-district differences in the rates of substantial assistance motions, with a de-fendant being approximately 9% more likely to be granted this form of downwarddeparture in more “lenient” districts. As previously noted, the application of a sub-stantial assistance motion is often applied “to avoid guideline ranges or mandatoryminimum sentences for sympathetic defendants - even when there has been no gen-uine substantial assistance.”177 However, inter-district differences in average ratesof mandatory minimums and rates of substantial assistance motions do not appearto have increased significantly following Booker and Kimbrough/Gall.

IV POLICY RECOMMENDATIONS

This section describes three of the major proposals for reform of federal sentencingafter Booker. I describe each in turn, and then apply the empirical findings in thispaper to shed light on the desirability of the various proposals, assuming that areduction in inter-judge disparities is a worthwhile goal of sentencing reform.

177Nagel & Schulhofer, supra note 37, at 550.


A. “Topless” Guidelines

Within a few months after Booker, the Department of Justice recommended a new“topless” Guidelines system, in which judges would be bound by the Guidelinesminimum, but not the maximum.178 Echoing the topless Guidelines regime firstproposed by Professor Frank Bowman, this construction would still allow judicialfact-finding to facts that raised the minimum applicable sentence, remaining con-stitutional under the principles espoused in Blakely.179 Recall that Blakely appliedthe Sixth Amendment to challenge judicial fact-finding which raised a defendant’smaximum sentence.180 As a result, the recommended “topless” Guidelines systemappeared to comport with both Blakely and the Court’s holding in Harris v. UnitedStates that facts triggering a mandatory minimum sentence could be found by ajudge.181 However, the constitutional viability of a “topless” Guidelines system hasnow been firmly rejected, with the Supreme Court’s recent holding in Alleyne v.United States, in which the Court squarely overruled Harris.182

Moreover, even if constitutionally permissible under the Sixth Amendment, the“topless” regime takes the prior mandatory Guidelines as the baseline, which someargue “would constitute a step backwards in the development and evolution of thefederal sentencing system by exacerbating some of the worst features of the pre-Booker federal sentencing.”183 By binding judges to the applicable minimum sen-tence, the proposal would likely re-introduce concerns associated with prosecutorialpower in charging and plea bargaining.184

Indeed, the evidence from Part IV provides empirical support for the propositionthat a “topless" Guidelines proposal would aggravate disparities that are attributableto prosecutorial charging decisions. Table 1 and Table 5, which present evidenceof inter-judge disparities and inter-district disparities, are reduced by almost a fac-tor of two when mandatory minimums are excluded from analyses. These resultssuggest that the decision to charge a mandatory minimum contributes substantiallyto inter-judge differences, such that these decisions are not made equally across

178See Federal Guidelines Sentencing Speech, supra note 23, at 326 (favoring “the constructionof a minimum guideline system”).

179See Memorandum from Frank Bowman to U.S. Sentencing Comm’n (June 27, 2004), 16 FED.SENTENCING REP. 364, 367 (2004).

180Blakely, 542 U.S. 296.181536 U.S. 545, 567-69 (2002). However, some commentators have suggested that the Court’s

holding in Harris may not survive after Booker. See, e.g., Erwin Chemerinsky, Making Sense ofApprendi and its Progeny, 37 MCGEORGE L. REV. 531, 541 (2006); Frank O. Bowman III, Mr.Madison Meets a Time Machine: The Political Science of Federal Sentencing Reform, 58 STAN. L.REV. 235, 261 (2005).

182Alleyne v. United States, 133 S. Ct. 420, slip op. at 15 (2012) (“Because there is no basis inprinciple or logic to distinguish facts that raise the maximum from those that increase the minimum,Harris was inconsistent with Apprendi. It is, accordingly, overruled.”).

183See Berman, Tweaking Booker, supra note 112, at 363. Berman also discusses potential con-stitutional challenges to a “topless” Guidelines system.

184Id. at 364 (“Consequently, the most problematic facets and the most disconcerting conse-quences in terms of prosecutorial power, disparity, and evasion experienced in the pre-Booker fed-eral sentencing system would likely be aggravated by the enactment of any sort of topless guidelineBooker fix.”).


all eligible cases. The results also indicate that mandatory minimum practices dif-fer largely across U.S. district courts. Accordingly, any proposal that binds judgesto the applicable minimum sentence would ascribe greater power to prosecutors,likely resulting in inequitable disparities. Furthermore, results in Table 6 suggestthat there have been substantial increases in inter-judge disparities in above rangedepartures even when a mandatory minimum is not charged. As a result, to theextent that a “topless” regime seeks to limit judicial discretion, it does so in anasymmetrical manner.

B. “Blakely-ized” Guidelines

Justice Breyer’s ultimate remedy for the Sixth Amendment issues facing the Fed-eral Sentencing Guidelines was to declare the Guidelines “effectively advisory.”But one could have imagined another approach: to “Blakely-ize" the Guidelines.Indeed, the Justices dissenting from the Breyer remedial opinion in Booker sug-gested leaving the mandatory Guidelines intact, but requiring that aggravating factstriggering longer maximums be proven by a jury beyond reasonable doubt, or ad-mitted by the defendant.185

However, introducing jury fact-finding into a mandatory Guidelines system islikely particularly complex. Justice Breyer in his Booker remedial opinion musedover how jury fact-finding might work, asking “[w]ould the indictment have to al-lege, in addition to the elements of robbery, whether the defendant possessed afirearm, whether he brandished or discharged it, whether he threatened death...?”186

Additionally, to the extent that the mandatory Guidelines regime contributed toprosecutorial discretion and disparity, jury fact-finding in the face of extensive pleabargaining “would move the system backwards in respect to both tried and plea-bargained cases” by effectively “prohibit[ing] the judge from basing a sentenceupon any conduct other than the conduct the prosecutor chose to charge.”187

Other scholars have echoed the concern that Blakely-izing the current version ofthe Guidelines would be procedurally unworkable and overwhelm juries requiredto make findings of fact.188 Addressing some of Justice Breyer’s concerns, a 2005American Bar Association Report suggested a version of the Blakely-ized systemespoused by Justice Stevens in his Booker dissent, accompanied with “simplifyingthe Guidelines by reducing both the number of offense levels and the number ofadjustments and presenting the remaining, more essential, culpability factors to the

185Booker, 543 U.S. at 284-84 (Stevens, J., dissenting) (“Rather than engage in a wholesale rewrit-ing of the SRA, I would simply allow the Government to continue doing what it has done since thisCourt handed down Blakely–prove any fact that is required to increase a defendant’s sentence underthe Guidelines to a jury beyond a reasonable doubt.” ).

186Id. at 254.187Id. at 256 (“plea bargaining would likely lead to sentences that gave greater weight, not to

real conduct, but rather to the skill of counsel, the policies of the prosecutor, the caseload, and otherfactors that vary from place to place, defendant to defendant, and crime to crime...plea bargainingof this kind would necessary move federal sentencing in the direction of diminished, not increased,uniformity in sentencing”). For a more thorough discussion of the potential problems with thisparticular recommendation, see Berman, Tweaking Booker, supra note 112, at 365-71.

188See Bowman, Beyond Band-Aids, supra note 109, at 191 (“the consensus view is that theGuidelines as now written are simply too complex and confusing to operate through juries”).


jury.”189

This Article cannot comment on the relative abilities of judges and juries todetermine the applicability of aggravating and mitigating factors. Even supposingthat juries are capable of fact determinations of aggravating and mitigating factors,under the “Blakely-ized” Guidelines, once a jury has made factual determinationsas to conduct based on what a prosecutor chose to charge, a judge is bound bythese determinations. For instance, if a jury did not make a factual determinationwith respect to a potential mitigating factor, such as acceptance of responsibilityby the defendant, a judge would not be allowed to consider this factor, even if itwere applicable. This Article provides evidence suggesting that a large componentof disparities stem from prosecutorial charging decisions. A requirement of juryfact-finding in a mandatory Guidelines regime may exacerbate such disparities.

C. Judge Sessions Proposal

Most recently, former Sentencing Commission Chair Judge William K. SessionsIII has recommended adoption of a simplified presumptive Guidelines system.190

Judge Sessions argues in favor of a new sentencing regime that balances two goals:(1) the need to reduce unwarranted sentencing disparities curbing the ability ofjudges to use subjective notions of justice to mete out punishment, and (2) givingjudges discretion to tailor sentencing to the unique circumstances of offenders andoffenses.191 Judge Sessions recommends a reduction in the number of possiblesentencing ranges, but broader ranges to afford judges greater discretion.192 In orderto comply with the constitutional requirements identified in Blakely, Judge Sessionssuggests that any facts that would increase the base offense level would have to beproven by a jury beyond a reasonable doubt, unless admitted to by the defendant.193

Judge Sessions also proposes simplifying the Guidelines by reducing the num-ber of aggravating or mitigating factors that increase or decrease the base offenselevel under Chapter Two and Chapter Three of the Guidelines Manual, which manyhave argued are overly complex.194 In deciding which aggravating factors to keepwithin the Guidelines, Sessions argues in favor of the strategy suggested by JusticeBreyer - empirically reviewing which enhancements in Chapter Two are commonlyused.195

189ABA Criminal Justice Section, Report and Recommendation on Booker (Jan. 2005), reprintedin 17 FED. SENT. REP. 335, 339 (2005).

190Sessions, supra note 24, at 340.191Id. at 339.192Id. at 340-45 (describing recommended changes to the current Guidelines sentencing chart).193Id. at 346.194Id. at 347-48; see also Frank O. Bowman III, The Failure of the Federal Sentencing Guide-

lines, 105 COLUM. L. REV. 1315, 1341 (2005); Stephen Breyer, Federal Sentencing GuidelinesRevisited, 11 FED. SENTENCING REP. 180 (1999) (“[T]he Guidelines are simply too long and toocomplicated.”).

195Sessions, supra note 24, at 349; Stephen Breyer, supra note 53, at 184 ("... I believe theCommission should review the present Guidelines, acting forcefully to diminish significantly thenumber of offense characteristics attached to individual crimes. The characteristics that remainshould be justified for the most part by data that shows their use by practicing judges to changesentences ... .").


Finally, Judge Sessions suggests a new form of appellate scrutiny because “[t]hethreat of reversal [on appeal] is a key component of [effective] guidelines,” withwithin range sentences “essentially unreviewable on appeal ... [unless] a districtcourt refused to consider all relevant factors or instead considered a prohibited fac-tor, such as a defendant’s race or gender.”196 In contrast, Judge Sessions proposes“relatively strict scrutiny by the appellate court” for downward departures.197

Critics of the Sessions proposal argue that the proposal would eliminate “ju-dicial feedback to the Commission and constructive evolution of the [G]uidelineswould virtually cease” as judges would have limited authority in setting the appli-cable sentence range.198 As a result, both scholars and district court judges haveexpressed the view that the current advisory Guidelines best achieves the goals ofsentencing because it reflects the right balance between various actors in federalsentencing.199 Of district judges surveyed in 2010, over 75% prefer the current ad-visory Guidelines system to other alternatives.200 In contrast, 14% of judges favoreda version of the Blakely-ized proposal - “[a] system of mandatory [G]uidelinesthat comply with the Sixth Amendment and have broader sentencing ranges thancurrently exist, coupled with fewer statutory mandatory minimums.”201 Only 3%of judges preferred a return to the pre-Booker Guidelines system, suggesting thatthe overwhelming majority of judges would be opposed to a return to presumptiveGuidelines, as proposed by Judge Sessions.202

Undoubtedly, Booker has given judges the freedom to consider the particularcircumstances of the offense and traits of the defendant. To the extent that grow-ing inter-judge disparities are reflective of these considerations, disparities are war-ranted and judicial discretion is desirable. On the other hand, some have suggestedthat the shift to advisory Guidelines has been accompanied by increases in unwar-ranted disparities.203

The empirical findings in this Article reveal that inter-judge disparities havedoubled from the period of mandatory Guidelines sentencing to post Booker sen-tencing, with a defendant potentially receiving a 5 month longer sentence due to themere happenstance of the judge assigned. Undoubtedly, a return to “presumptive”Guidelines would mechanically reduce inter-judge disparities by greatly limiting

196Sessions, supra note 24, at 353-54.197Id. at 353-54 (“District courts’ choices of sentences within the applicable cells on the grid

would be essentially unreviewable on appeal so long as the courts considered all of the relevantaggravating and mitigating factors identified in the application notes and all other relevant factors inthe Guidelines Manual before imposing a particular sentence.”).

198Baron-Evans & Stith, supra note 25, at 1716.199Id. at 1681. See also Michael Tonry, The U.S. Sentencing Commission’s Best Response to

Booker is to Do Nothing, 24 FED. SENT. REP. 387 (2012); Sara Sun Beale, Is Now the Time forMajor Sentencing Reform?, 24 FED. SENT. REP. 382 (2012).

200See U.S. Sentencing Commission, RESULTS OF SURVEY OF UNITED STATES DISTRICTJUDGES JANUARY 2010 THROUGH MARCH 2010 (June 2010), at 23 (Question 19, Table 19).

201Id.202Id.203Sessions, At the Crossroads, supra note 24; Bowman, Nothing is Not Enough, 24 FED. SENT.

REP. 356 (June 2012) (“[T]he post-Booker advisory system retains most of the flaws of the systemit replaced, while adding new ones, and its sole relative advantage - that of conferring additional(and effectively unreviewable) discretion on sentencing judges - is insufficient to justify its retentionas a permanent system.”).


judicial discretion. However, the empirical evidence from Part IV seeks to ascer-tain the effect of the sentencing regime on inter-judge disparities in outcomes thatare most likely attributable to judge behavior. Differences in sentence lengths canbe attributable to both judge disparities as well as differences in charging of manda-tory minimums. The findings in this Article suggest that a return to a presumptiveregime, without any changes in mandatory minimums, would only go partway inreducing disparities, and curtail potentially desirable judicial discretion.

While this Article does not provide evidence supporting a return to “presump-tive” Guidelines, it does suggest that strictness of appellate review is a potentiallyimportant constraint on judicial discretion in sentencing. Inter-judge disparities inbelow range departures were generally lowest during the PROTECT Act, which im-posed de novo review. Furthermore, empirical evidence suggests that Booker alonedid not contribute to recent increases in inter-judge disparities. In the first two yearsafter Booker, inter-judge disparities were not statistically different from that duringKoon. Rather, it appears to be the impact of Booker plus reduced appellate scrutinyfollowing Rita, Gall and Kimbrough that are responsible for increases in inter-judgedisparities.

Thus, reforms to strengthen the degree of appellate review could possibly re-duce inter-judge sentencing disparities. In Gall, the Court did not require appellatecourts to insist upon “extraordinary” circumstances to justify a sentence outsidethe Guidelines recommended range, specifically rejecting stronger justifications forsentences that departed more greatly from the Guidelines.204 In order to constraininter-judge disparities, the Commission could require district court judges to pro-vide a heightened justification for more severe departures from the prescribed sen-tence, without coming too close to an “ impermissible presumption of unreason-ableness for sentences outside the Guidelines range.....[which] would not be con-sistent with Booker.”205

V CONCLUSION

Exploiting the random assignment of cases to judges in district courthouses repre-senting 73 U.S. district courts, this Article finds a significant increase in inter-judgedisparities from the Koon period to after Kimbrough/Gall. A defendant sentencedby a “harsh” judge prior to Koon was sentenced to 2.6 months longer than the av-erage, but over 5 months longer after Kimbrough/Gall, a doubling of inter-judgedisparities. Increased inter-judge disparities persist even excluding cases in whichmandatory minimums were charged, suggesting that judges are not completely an-chored to the Guidelines.

Increases in between-judge differences following Booker and Kimbrough/Gallappear to be linked to observable judicial demographics such as gender, political

204Gall, 552 U.S. at 47 (“In reviewing the reasonableness of a sentence outside the Guidelinesrange, appellate courts may therefore take the degree of variance into account and consider the ex-tent of a deviation from the Guidelines. We reject, however, an appellate rule that requires "extraor-dinary" circumstances to justify a sentence outside the Guidelines range. We also reject the use of arigid mathematical formula that uses the percentage of a departure as the standard for determiningthe strength of the justifications required for a specific sentence.”).

205Id. at 47.


affiliation of appointing president, and whether a judge has ever sentenced underthe mandatory Guidelines regime. I also find modest evidence of increases in inter-district differences following Kimbrough/Gall, with large inter-district differencesin sentence length, rates of below range departures, rates of mandatory minimums,and rates of substantial assistance motions. However, the magnitudes of both inter-judge and inter-district disparities are drastically smaller when mandatory mini-mums are excluded, suggesting that prosecutorial charging decisions are a majorcontributor to sentencing disparities.

Overall, these results suggest that the shift to an advisory Guidelines regimeunder Booker, coupled with lowered standards of appellate scrutiny, have led togreater inter-judge disparities. Prosecutorial charging decisions, at least in the ap-plication of mandatory minimums, appear to play a substantial role in explainingdisparities. While a first step in disentangling the sources of disparities ascribableto various actors, a primary limitation of this Article is its inability to thoroughlyanalyze all the disparities that can arise in earlier stages of the criminal justice sys-tems, such as through charging and plea bargaining. Nevertheless, the results of thisArticle caution against recent proposals to move back towards a sentencing systemin which judges are bound by the decisions of prosecutors. Instead, the Article sug-gests that it may be wise to modify standards of appellate review, as well as revisitthe desirability of mandatory minimums.


APPENDIX

A. Testing for Random Assignment

To test for random assignment, I regress five defendant characteristics on districtcourthouse by sentencing year fixed effects, sentencing month fixed effects andjudge fixed effects. The five defendant characteristics include: gender, age, a blackrace indicator, number of dependents, and an indicator for less than a high schooldegree. Intuitively, there should be no significant correlation between a particularjudge and defendant characteristics if cases are randomly assigned.

However, in testing the random assignment of defendants across these five char-acteristics, I encounter the problem that defendant characteristics are not fully in-dependent. For instance, black defendants are also likely to have completed lessthan a high school degree. To address the confounding nature of these character-istics, I use seemingly unrelated regression (SUR) to test for random assignment.SUR allows me to test random assignment simultaneously for all the five defendantcharacteristics, addressing correlations.206

SUR can be formally described as the regression model:

Yijdtm = α0 + γd + δt + γd ∗ δt + λm + κj + εijdtm

where Yijdtm is a characteristic of defendant i, sentenced by judge j in district courtd in year t and month m. The specification includes district court fixed effects(γd), sentencing year fixed effects (δt), sentencing month fixed effects (λm), andsentencing judge fixed effects( κj) to accurately compare cases assigned to judgesin the same courthouse, and in the same year and month.

To formally test for random assignment, I test the null hypothesis of no judgeeffects - κ - using an F-test. The p-value for this F-test tests whether the defendantcharacteristics do not differ significantly among the cases that are assigned to dis-trict court judges in the same district courthouse, sentencing year, and sentencingmonth. A large p-value would signify the acceptance of the null hypothesis, andlead to the conclusion that random assignment was present.

See Table A1 for the randomization checks done by district courthouse, alongwith associated p-values. I drop all courthouses with F-test p-values less than 0.05,but results are robust to other cutoffs. Dropped courthouses are indicated with **.

206Testing each characteristic individually would result in incorrect standard errors if the demo-graphic characteristics are correlated. For a discussion of the SUR technique, see David H. Autor& Susan N. Houseman, Do Temporary-Help Jobs Improve Labor Market Outcomes for Low-SkilledWorkers? Evidence from “Work First”, 2 AEJ: Applied Economics 96, 106-107 (2010).


TABLE A1. RANDOMIZATION TESTS 2000-2009District Court No. Obs. P-valueME (0) 1,668 0.1438MA (1) 4,042 0.1054NH (2) 1,617 0.9844PR (4) 6520 0.2674CT** (5) 664 0.0000NY North - Syracuse (6) 1,148 0.1074NY East** (7) 12,447 0.0004NY South - White Plains (8) 1,338 0.4336NY West - Rochester (9) 1,166 0.6226VT (10) 1,400 0.2379DE (11) 641 0.3831NJ -Trenton (12) 476 0.2983PA East** (13) 6,411 0.0000PA Middle - Scranton (14) 969 0.6837PA Middle - Williamsport (14) 234 0.2071PA West - Erie (15) 609 0.0521PA West - Pittsburgh (15) 2,917 0.0645MD (16) 5,569 0.0631NC East - Southern (17) 608 0.3847NC Middle (18) 3,205 0.08086NC West**(19) 5,563 0.0000SC** (20) 8,848 0.0000VA East -Alexandria (22) 4,500 0.3178VA East -Norfolk (22) 1,105 0.1658VA East -Newport News (22) 743 0.0662VA West (23) 3,123 0.3250WV North - Martinsburg (24) 639 0.4091WV South (25) 1,778 0.0932AL North** (26) 1,430 0.0189AL Middle (27) 904 0.3242AL South (28) 3,132 0.0702FL North (29) 2,718 0.5783FL Middle - Ft. Myers (30) 923 0.3824FL Middle - Ocala (30) 465 0.3128FL South - Ft. Pierce (31) 3,299 0.0541FL South - Ft. Lauderdale (31) 649 0.2485GA North** (32) 5,823 0.0000GA Middle (33) 2,064 0.1396LA East (35) 3,117 0.0606LA West (36) 1,686 .6360MS North (37) 925 0.4247MS South (38) 3,057 0.0564


TABLE A1. RANDOMIZATION TESTS 2000-2009 (CONTINUED)TX North - Forth Worth (39) 2,027 0.2386TX East (40) 6,563 0.5598TX South - Brownsville (41) 10,112 0.3364TX South - Corpus Christi (41) 6,679 0.2767TX South - Laredo (41) 19,079 0.6244TX South - McAllen (41) 12,739 0.1093TX West - Del Rio (42) 7,098 0.3500TX West - Midland-Odessa (42) 3,567 0.4120KY East - Covington (43) 717 0.5872KY East - Pikeville (43) 139 0.0966KY East - Lexington (43) 1,993 0.8694KY West (44) 1,746 0.1114MI East - Bay City (45) 458 0.4009MI East - Flint (45) 673 0.3014MI West (46) 3,313 0.0961OH North - Toledo (47) 1,014 0.2105OH South - Dayton (48) 1,300 0.9115TN East (49) 5,200 0.0705TN Middle** (50) 1,938 0.0126TN West - Eastern (51) 831 0.3998IL North - Rockford (52) 624 0.8929IL Central (53) 2,618 0.1283IL South (54) 2,736 0.1296IN North - South Bend (55) 954 0.2764IN North - Fort Wayne (55) 530 0.0741IN South (56) 2,004 0.3266WI East - Milwaukee (57) 2,206 0.4223WI West (58) 1,571 0.1123AR East (60) 2,739 0.1631AR West**(61) 1,098 0.0001IA North (62) 2,413 0.0561IA South (63) 2,684 0.8265MN** (64) 4,815 0.0001MO East (65) 8,203 0.0785MO West (66) 6,764 0.1191NE - Omaha (67) 2,323 0.0532ND (68) 1,888 0.2250SD - Aberdeen (69) 309 0.1479SD - Pierre (69) 1,010 0.8757AZ - Tuscon (70) 23,677 0.0961AZ - Yuma (70) 2,449 0.3392CA North (71) 3,045 0.1970CA East (72) 8,094 0.0646


TABLE A1. RANDOMIZATION TESTS 2000-2009 (CONTINUED)CA Central - Riverside (73) 157 0.4520CA South - El Centro (74) 8,664 0.3442CA South - Yuma (74) 89 0.3502HI** (75) 3,351 0.0012ID (76) 1,526 0.0544MT - Missoula (77) 516 0.1698MT - Great Falls (77) 1,003 0.2206NV (78) 4,867 0.6549OR - Eugene (79) 954 0.2261OR - Medford (79) 434 0.6618WA East - Spokane (80) 1,401 0.3100WA West** (81) 5,302 0.0001CO** (82) 4,582 0.0000KS (83) 5,509 0.2031NM (84) 24,019 0.2924OK North (85) 1,279 0.3240OK East (86) 736 0.9312OK West** (87) 1,809 0.0001UT (88) 5,276 0.9421WY** (89) 1,565 0.0002DC (90) 346 0.5720AK (95) 1,218 0.1105LA Middle** (96) 1,112 0.0263

Notes: Data is from the random sample from 2000-2009. I drop judges who retired or were terminated prior to2000, and judges and district offices with an annual caseload of less than 25. For each district court, I controlfor district office by sentencing year, sentencing month, and judge fixed effects. P-values reported test whetherjudge fixed effects differ significantly from zero and are from a seemingly unrelated regression (SUR) on fivedefendant characteristics: defendant gender, age, black race indicator, number of dependents, and less than highschool indicator. ** indicates dropped courthouses.

TABLE A2. INTER-DISTRICT VARIATION

APPLICATION OF MANDATORY MINIMUMS

Period σ Lower bound Upper bound No. Obs.Koon .0571 .0491 .0663 183732PROTECT Act .0652 .0559 .0760 94434Booker .0584 .0502 .0680 165139Kimbrough/Gall .0662 .0568 .0771 117536Notes: Data is from the full sample sample from 2000-2009. All regressions contain demographiccontrols, and controls for offense type, offense level and criminal history. Regressions also containsentencing year fixed effects, and sentencing month fixed effects.


TABLE A3. INTER-DISTRICT VARIATION

APPLICATION OF SUBSTANTIAL ASSISTANCE

Period σ Lower bound Upper bound No. Obs.Koon .0868 .0747 .1009 176736PROTECT Act .0909 .0782 .1058 92736Booker .0904 .0779 .1049 163978Kimbrough/Gall .0789 .0679 .0918 117328Notes: Data is from the full sample sample from 2000-2009. All regressions contain demographiccontrols, and controls for offense type, offense level and criminal history. Regressions also containsentencing year fixed effects, and sentencing month fixed effects.

B. Analysis of Variance

I implement an analysis of variance using a defendant-level random effects specifi-cation of the form:

Yijdtm = Xi ∗ β + γd + δt + γd ∗ δt + λm + vijdtm,

where vijdtm = µjdtm + εijdtm

The dependent variable Yijdtm is the sentence length in months for defendant iassigned to judge j in district court d, sentenced in year t and month m. The controlvariables include defendant and crime characteristics (Xi), sentencing year fixedeffects (δt), and sentencing month fixed effects (λm). γd are indicator variablesfor the district courthouse in which sentencing occurred. The residual vijdtm iscomposed of a judge effect or value added that is constant for a judge over time, andan idiosyncratic defendant effect. I estimate the coefficients β and the judge effectsµ by ordinary least squares (OLS) regression. OLS estimation yields consistentestimates of β if the judge random effects are uncorrelated with the control variablesX .

I estimate the magnitude of the judge effects under a mixed random effectsspecification, assuming that µjdtm is distributed N(0, σ2

µ). Intuitively, within judgevariance in vijdtm is used to estimate the defendant variance:

σ̂2ε = V ar(vijdtm − v̄jdtm)

The variance in the judge effect is the remainder:

σ̂2µ = V ar(vijdtm)− σ̂2

ε

The estimated standard deviation of judge effects on sentence length is σµ = X,implying that a one standard deviation increase in judicial harshness raises a defen-dant’s sentence by X months. Because the regression specification includes districtcourthouse fixed effects, this measure represents the impact of being assigned ajudge one standard deviation higher in harshness in the within district court distri-bution.


C. Judge Demographic Regression

To analyze the differential sentencing practices of certain types of judges, I use OLSregression. The methodology captures how judges differ in their treatment of simi-lar defendants in response to increased judicial discretion, compared to other judgeswithin the same district courthouse. Because cases are randomly assigned to judgeswithin a district court, judge identifiers allow one to compare judges within the samecourt, capturing judge differences in sentencing rather than different caseloads.

I identify the sources of increasing inter-disparities post Booker and post Kim-brough/Gall using a specification of the form:

Yicodtm = β0 + α1 ∗ Judgei ∗Booker + β0 ∗Booker + α2 ∗ Judgei ∗Kimbrough+β0 ∗Kimbrough+ β1 ∗Racei + β3 ∗Xi +Guideico +Offtypei

+γd + δt + γd ∗ δt + λm + εicodtm

Yicodtm is a sentencing outcome for defendant i, with criminal history category c andoffense level o, sentenced in district court d in year t and month m. Main outcomesinclude sentence length measured in months, and binary indicators for below rangesentencing and above range sentencing.

Judgei includes judicial demographics such as race, gender, political affiliation,an indicator for pre vs. post Guidelines appointment, tenure under the Guidelines,and an indicator for pre vs. post Booker appointment. The main coefficients α1 andα2 capture the impact of particular judicial characteristics on sentencing outcomesin the wake of Booker and its progeny. Booker is an indicator variable for defen-dants sentenced after the Booker decision but before Kimbrough/Gall. Kimbroughis an indicator variable for defendants sentenced after Kimbrough/Gall.

Racei is a dummy variable for defendant i’s race: white, black, Hispanic, orother. Xi comprises a vector of demographic characteristics of the defendant in-cluding gender, age, age squared, educational attainment (less than high school,high school graduate, some college, college graduate), number of dependents, andcitizenship status.

Guideico includes dummy variables for criminal history category c and offenselevel o, and each unique combination of criminal history category and offense level.The interaction captures differential sentencing tendencies at each unique cell ofthe Guidelines grid (258 total). To proxy for underlying offense seriousness and allaggravating and mitigating factors, I control for final offense level. I also controlfor final criminal history category. Offtypei is a dummy variable for offense type.

The specification also includes district court fixed effects (γd), sentencing yearfixed effects (δt), and sentencing month fixed effects (λm). All standard errors areclustered at the district courthouse level to account for serial correlation.

Documents

Have Inter-Judge Sentencing Disparities Increased in an Advisory